Consumption Notes

Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
In Principles of Macroeconomics, and perhaps Intermediate Macroeconomics, you were introduced to
the Keynesian consumption function:
C = a + bY
where C is consumption, Y is income, and a and b are parameters.
This equation is obviously a reduced form relationship. However, there are many instances in which
economists and policymakers might be interested in the marginal propensity to consume (b in the equa-
tion above). As a result, we need a theory of consumption.
Thus, in this course, we want to move beyond this basic consumption equation and think about the
consumption decision in a more rigorous way. Specifically, we are going to analyze the consumption
decision of households in the same way that we analyze microeconomic decisions. However, this analysis
will have one notable exception compared to standard microeconomic decisions in that it will involve
time. For example, in microeconomics, you often had a consumer who chose the quantities of n different
goods to maximize utility given a budget constraint. Here, we will have a household that is choosing
baskets of consumption over time to maximize utility subject to a budget constraint.
We will begin with an infinite-horizon household problem. We will show how to solve the model in
discrete time using Lagrange’s method and dynamic programming. We will also show how to solve the
model in continuous time using the maximum principle. We also discuss the intuition of the infinite-
horizon model, how we can use the model to illustrate the permanent income hypothesis, and how we
can relate the permanent income hypothesis to the Keynesian consumption function. We will also discuss
the model of perpetual youth, in which there is some probability that the consumer dies each period.
Finally, we will consider the role of portfolio allocation for consumption when there are two assets and
one is risky.

2 An Infinite Horizon-Consumption Model

2.1 Discrete Time
Time is discrete and continues forever. We will focus on the consumption decision of an infinitely-lived
household. In the previous period, the household had assets balances, at . The household enters the
present period with asset balances, (1 + r)at , where r is the real interest rate and is assumed to be
constant. Each period, the household receives an endowment of income, xt . The household’s budget
constraint is therefore
at+1 + ct = xt + (1 + r)at (1)
where ct is consumption.

1
 Suppose that the household has a utility function, u(ct ). The expected lifetime utility of the household
is
∞
X
U = E0 β s u(ct+s ) (2)
s=0

where β is the discount factor.1

The household wants to maximize its utility. However, think about the problem facing the household.
In period t, the household wants to choose {ct , at+1 } to maximize U (ct ), given xt , at , and r. In period
t + 1, the household wants to choose {ct+1 , at+2 } to maximize utility. One way to think about this is to
think about the household as choosing the entire future path of consumption at time t. To illustrate this
decision, we will use two approaches. In the first approach, we will set up a Langrangian for an infinite
horizon. The second approach we use is that of dynamic programming.

2.1.1 Solving the Household Problem with a Lagrangian

We can write the Langrangian for the household’s problem as
∞ 
X 
s
max L= β u(ct+s ) + λt+s [xt+s + (1 + r)at+s − ct+s − at+s+1 ]
ct+s ,at+s ,λt+s
s=0

The first-order conditions for maximization are

β s u0 (ct+s ) = λt+s ∀s

(1 + r)λt+s = λt+s−1 ∀s
Now let’s combine the first-order conditions. Also, since these conditions hold for all values of s, let’s
use s = 1.
u0 (ct ) = β(1 + r)u0 (ct+1 ) (3)
This equation is what is known as the consumption Euler equation.2

2.1.2 Solving the Household Problem with Dynamic Programming

We can also solve this problem using dynamic programming. The basic idea is as follows. In the decision
faced by the household, there is a state variable and a control variable. The budget constraint describes
the evolution of the state variable, at . The household chooses the control variable, ct , to maximize its
objective function. This choice then determines the value of the state next period, at+1 . Given the value
of that state variable, the household then chooses ct+1 to maximize utility in the next period, which in
turn determines at+2 . Thus, the maximization problem of the household is really just the same as solving
successive static optimization problems. Given the choice at time t, the decision at time t + 1 should
follow optimally from that initial choice and so on.
Formally, we can write the household’s problem using a Bellman equation:

V (at ) = u(ct ) + βEt V (at+1 )

where V (a) denotes the value of holding asset balances a, β is the discount factor, and Et is the
mathematical expectations operator.3 In addition, the path of the state variable is given by equation (1).
1 1
β = 1+ρ , where ρ is the pure rate of time preference.
2
More on this below.
3
The use of the expectations operator implies that there is uncertainty. If we assume that there is no uncertainty over the
household’s income, we can dispense with this notation.

2
 The household’s problem is therefore given as
 
V (at ) = max u(ct ) + βEt V [xt + (1 + r)at − ct ] (4)
ct

The first-order condition is given as

u0 (ct ) = βEt V 0 (at+1 )
Now differentiate the Bellman equation with respect to at :

V 0 (at ) = (1 + r)βEt V 0 (at+1 )

Using the envelope theorem,

V 0 (at ) = (1 + r)u0 (ct )
It follows that
V 0 (at+1 ) = (1 + r)u0 (ct+1 )
Substituting this expression into the first-order condition yields

u0 (ct ) = β(1 + r)Et u0 (ct+1 ) (5)

Note that this is just our consumption Euler equation shown in equation (3).

2.1.3 The Permanent Income Hypothesis

As alluded to above, Keynes’s basic consumption function has been supplanted in the literature by the
permanent income hypothesis.4 We will now solve the household problem in a different way to illustrate
the permanent income hypothesis.
Suppose that one has a function, f (x). A first-order Taylor series approximation around a is given
as
f (a) + f 0 (a)f (x − a)
Given the household utility function u(ct+1 ), a first-order Taylor series approximation around ct can
be written as
u0 (ct+1 ) ≈ u0 (ct ) + u00 (ct )(ct+1 − ct )
Dividing this expression by u0 (ct ) yields

u0 (ct+1 ) u00 (ct )

= 1 + ∆ct+1
u0 (ct ) u0 (ct )

Now recall our Euler equation (3):

u0 (ct ) = β(1 + r)u0 (ct+1 )
It follows that
u00 (ct ) 1
1+ 0
∆ct+1 =
u (ct ) β(1 + r)
4
The permanent income hypothesis was proposed by Milton Friedman in a book entitled, A Theory of the Consumption
Function. Franco Modigliani proposed a similar hypothesis known as the life-cycle hypothesis. Both men later won the Nobel
Prize, in part for this work. Given the similarities, this is sometimes referred to as the Permanent Income-Life Cycle Hypothesis
(PILCH).

3
Given the definition of β, this can be written as

u00 (ct ) ρ−r

0
∆ct+1 =
u (ct ) 1+r

where ρ is the pure rate of time preference.

Let u(ct ) = c1−σ
t /(1 − σ). Then  
∆ct+1 1 r−ρ
=
ct σ 1+r
The intuition behind this expression is as follows. Suppose that the real interest rate is greater than the
rate of time preference. In this case, the change in consumption from time t to time t + 1 would be
positive since the household is pushing its consumption more into the future (saving). On the other hand,
if the real interest rate is lower than the rate of time preference, then the real interest rate is not enough
to compensate the household to forgo current consumption and the growth rate of consumption will be
negative (less saving/more borrowing).
We need to think about what is going on in general equilibrium. In the long run, r = ρ in equilibrium.
However, this implies that ∆ct+1 = 0. Put differently, in the long run, ct = ct+1 = ct+2 = . . . .
Consumption is constant over time.
To see why this result matters, let’s recall the budget constraint from the household’s problem:

at+1 + ct = xt + (1 + r)at

Update this one period,

at+2 + ct+1 = xt+1 + (1 + r)at+1
Now substitute this into the first equation to eliminate at+1 ,
at+2 ct+1 xt+1
+ ct + = xt + + (1 + r)at
1+r 1+r 1+r
Continuing this substitution process n times and then taking the limit as n → ∞ will ultimately result in
∞ ∞
X ct+s X xt+s
= + (1 + r)at
(1 + r)s (1 + r)s
s=0 s=0

Note that if consumption is constant over time, this can be re-written as

∞ ∞
X 1 X xt+s
ct = + (1 + r)at
(1 + r)s (1 + r)s
s=0 s=0

We can simplify this further.

Define
1
as =
(1 + r)s
It follows that
∞
X
Z= as = 1 + a + a2 + a3 + . . .
s=0

aZ = a + a2 + a3 + a4 + . . .
Z − aZ = 1

4
 1
Z=
1−a
1
a=
1+r
∞ ∞
X
s
X 1 1 1 1+r
Z= a = = = 1 =
(1 + r)s 1−a 1 − 1+r r
s=0 s=0
Plugging this expression into the budget constraint:
∞
X 
xt+s
ct = r + at = rWt (6)
(1 + r)s+1
s=0

where Wt is wealth. Thus, consumption today is determined by the present discounted value of all future
income as well as current asset balances. In other words, consumption is a function of household wealth.

2.1.4 Implications of PIH for Consumption Function

Recall that the Keynesian consumption function implied that consumption was a function of current
income. Suppose that you wanted to estimate the marginal propensity to consume, b, from the Keynesian
consumption function.
C = a + bY
Milton Friedman provided interesting ways to test the permanent income hypothesis given this approach.
To illustrate the basic concept, assume that you had panel data and you were going to estimate a
regression of the form
ci = a + byi + ei (7)
Estimating the consumption equation above by least squares would give you an estimate of b:
cov(c, y)
b̂ =
var(y)
where the subscripts have been neglected for expositional convenience.
Suppose that we had enough information to determine the fraction of an individual’s income that is
permanent and the fraction that is temporary and that each component of income is uncorrelated with
the other. Define total income as the sum of permanent and transitory income

y = yt + yp

The estimate of b could now be re-written as

cov(c, y p + y t ) cov(c, y p + y t )
b̂ = =
var(y p + y t ) var(y p ) + var(y t )
Assuming that consumption is just proportional to permanent income, c = y p , then
cov(y p , y p + y t ) var(y p )
b̂ = =
var(y p ) + var(y t ) var(y p ) + var(y t )
The the marginal propensity to consume is just a function of the variance of the permanent and transitory
components of income. In addition, one can generate empirical predictions based on this result. For
example, if one estimated the equation (7) for a cross-section of individuals, one would expect that people
who work in professions with large fluctuations in temporary income, such as real estate or farming, would
have lower marginal propensity’s to consume out of current income.

5
2.2 Continuous Time
Let’s now consider the household problem in continuous time. Let the household utility function be
given as Z ∞
U= e−ρt u[c(t)]dt
0
We can now write the household budget constraint as

ȧ = x + ra − c

2.2.1 The Maximum Principle

We can solve the household problem using the maximum principle. We can write the current-value
Hamiltonian as
H(c, a, λ) = u(c) + λ[x + ra − c]
where c is the control variable, a is the state variable, and λ is the co-state variable.
The conditions to satisfy the maximum principle are given as

u0 (c) = λ

ρλ − λ̇ = rλ
ȧ = x + ra − c
We can re-write the second condition as

λ̇
=ρ−r
λ
Let u(c) = c1−σ /(1 − σ). It follows that u0 (c) = c−σ . So the first condition for maximization can be
written as
c−σ = λ
Differentiating this expression with respect to time yields

−σc−σ−1 ċ = λ̇

Thus,
σc−σ−1 ċ
=r−ρ
c−σ
Or
ċ 1
= (r − ρ)
c σ
Note the similarity to the discrete case.

6
2.2.2 Using the Bellman Equation in Continuous Time
Note that in the discrete time environment, we had a Bellman equation of the form

V (at ) = max u(ct ) + βEV (at+1 )

ct

We can get an analogous equation in continuous time. Recall that we defined β = 1/(1 + ρ), where ρ is
the rate of time preference. Suppose that we think about this problem over time increment of time ∆t.
Suppose that the state variable a changes to some new value a0 during this increment of time. We can
write the Bellman equation as
1
V (a, t) = max u(c)∆t + EV (a0 , t + ∆t)
c 1 + ρ∆t
Multiply both sides by (1 + ρ∆t):

ρ∆tV (a, t) = (1 + ρ∆t) max u(c)∆t + EV (a0 , t + ∆t) − V (a, t)

c

Divide both sides by ∆t:

1
ρV (a, t) = (1 + ρ∆t) max u(c) + EV (a0 , t + ∆t) − V (a, t)
c ∆t
Take the limit of this expression as ∆t → 0. It follows that
1
ρV (a) = max u(c) + EdV
c dt
where lim∆t→0 [V (a0 , t+dt)−V (a, t)] = dV We can now use this Bellman equation to solve the household
problem. Note that
dV = V 0 (a)da
Recall from the budget constraint above that:

da = (x + ra − c)dt

It follows that the household’s problem is:

ρV (a) = max u(c) + (x + ra − c)V 0 (a)

c

The first-order condition is given as

u0 (c) = V 0 (a)
Let’s go back to our Bellman equation and differentiate with respect to a:
da
ρV 0 (a) = rV 0 (a) + V 00 (a)
dt
Let’s use the envelope theorem. From the first-order condition, we have

u00 (c)ċ = V 00 (a)ȧ

Plugging this back into the differentiated equation gives us

ρu0 (c) = ru0 (c) + u00 (c)ċ

7
Or,
u00 (c)
− ċ = r − ρ
u0 (c)
Suppose that
c1−σ
u(c) =
1−σ
Then,
u0 (c) = c−σ
u00 (c) = −σc−σ−1
Using these conditions yields
ċ 1
= (r − ρ)
c σ
Note that this is identical to the equation we found using the Maximum Principle.

3 A Model of Perpetual Youth

In the previous model we considered the consumption behavior of the household and assumed that the
household lived forever. Now we want to consider what happens if we think about an individual who
faces some probability of death. Suppose that the representative individual faces a probability of death,
p.
We can write the Bellman equation for this individual as

V (at ) = u(ct ) + (1 − p)βEt V (at+1 )

The individual’s budget constraint is the same as was assumed in equation (1). Thus, the individual’s
problem can be written as
 
V (at ) = max u(ct ) + (1 − p)βEt V [xt + (1 + r)at − ct ] (8)
ct

The first-order condition is given as

u0 (ct ) = (1 − p)βEt V 0 (at+1 )

Differentiating the Bellman equation with respect to a yields

V 0 (at ) = (1 + r)(1 − p)βEt V 0 (at+1 )

Using the envelope theorem,

V 0 (at ) = (1 + r)u0 (ct )
Thus, iterating forward,
V 0 (at+1 ) = (1 + r)u0 (ct+1 )
Substituting this into the first-order condition yields

Et u0 (ct+1 ) 1+ρ
0
= (9)
u (ct ) (1 − p)(1 + r)

Note that this is the standard Euler equation shown in (3), with the exception of the term (1 − p).

8
4 A Model of Consumption and Portfolio Allocation: Merton’s Model
In the basic consumption models above, we assumed that there was one riskless asset. The individual
chose consumption to maximize utility. Given some level of income and the consumption choice, the
residual is the individual’s savings. However, in the real world, there are both risky and riskless assets.
As a result, the individual must choose both consumption and how to allocate his/her portfolio. This is
precisely the model presented by Merton (1969, 1971).
Suppose an individual has wealth, w. the individual can allocate wealth between two different assets,
risk-free bonds and a risky asset. Suppose that the risk-free bond offers a rate of return r with absolute
certainty. Suppose the individual can hold another asset, a, which has a price, p, that follows a geometric
Brownian motion:
dpa
= µdt + σdz (10)
pa
where µ is the expected rate of change in √ the price, σ is the condition standard deviation, and dz is an
increment of a Wiener process (i.e. dz =  dt, where  is drawn from a standard normal distribution).
It follows that wealth can be written as

w = pa a + pb b

where b is the quantity of risk-free bonds and

dpb
= rdt
pb
Note that since the individual wants to use his/her wealth to consume, the evolution of wealth over time
is the change in the value of the individual’s assets less his/her consumption

dw = d(pa a + pb b) − cdt

where c is consumption and is assumed to be constant over the interval [t, t + dt). Note that

d(pa a + pb b) = adpa + pa da + bdpb + pb db = adpa + bdpb = µpa adt + σpa adz + rpb bdt

since b and a remain unchanged over the interval, [t, t + dt). Define s := (pa a)/w as the risky-asset
share of wealth. It follows that

dw = [µsw + r(1 − s)w − c]dt + σswdz (11)

The Bellman equation for the individual is given as

1
ρV (w) = max u(c) + EdV (12)
c,s dt
Using Ito’s Lemma, this equation can be written as
 
1 0 1 00
ρV (w) = max u(c) + E V (w)dw + V (w)(dw) 2
(13)
c,θ dt 2

From equation (11), it follows that

1
Edw = µsw + r(1 − s)w − c
dt

9
 1
E(dw)2 = (σsw)2
dt
Plugging this into equation (13) yields
1
ρV (w) = max u(c) + [µsw + r(1 − s)w − c]V 0 (w) + (σsw)2 V 00 (w)
c,s 2
Or, by simplifying,
1
ρV (w) = max u(c) + [rw + (µ − r)sw − c]V 0 (w) + (σsw)2 V 00 (w) (14)
c,s 2
The first-order conditions are
u0 (c) = V 0 (w)
(µ − r)V 0 (w) = −σ 2 swV 00 (w)
To get a solution, we need to know the value function. Suppose that the utility function is

c1−γ
u(c) =
1−γ
And guess that
w1−γ
V (w) = α−γ
1−γ
It follows that
u0 (c) = c−γ
V 0 (w) = (αw)−γ
V 00 (w) = −γα−γ w−γ−1
Plug these into the first order conditions:
c = αw
µ−r
s∗ = (15)
γσ 2
In the first equation, we have this pesky α term. Ugh. Let’s substitute the solution for c as well as our
guess for V (w) into the Bellman equation:

w1−γ (αw)1−γ 1
ρα−γ = + [rw1−γ + (µ − r)sw1−γ − αw1−γ ]α−γ − σ 2 s2 w1−γ γα−γ
1−γ 1−γ 2

Divide both sides by α−γ w1−γ . This yields

ρ α 1
= + r + (µ − r)s − α − γσ 2 s2
1−γ 1−γ 2
Solving for α yields
γ (µ − r)2
    
ρ γ−1
α= − r+ 1−
γ γ 2 σ2
Thus, the first-order condition for consumption can be re-written as

γ (µ − r)2
     
ρ γ−1
c= − r+ 1− (16)
γ γ 2 σ2

10
So consumption is proportional to wealth. Now consider the following special case: limγ→1 u(c) = ln(c).
In this case, α = ρ. It follows that the equilibrium conditions are given as

c = ρw (17)
µ−r
s= (18)
σ2
Thus, with log-utility, an individual’s consumption is equal to the product of the rate of time preference
and their wealth (note the similarity to section 2.1.3 above). And the fraction of their portfolio they devote
to the risky asset is determined by the Sharpe ratio.

References
[1] Friedman, Milton. 1957. A Theory of the Consumption Function. Princeton.

[2] Merton, Robert C. 1969. “Lifetime Portfolio Selection under Uncertainty: The Continuous-Time
Case.” The Review of Economics and Statistics, Vol. 51, No. 3, p. 247 - 257.

[3] Merton, Robert C. 1971. “Optimum Consumption and Portfolio Rules in a Continuous Time Model.”
Journal of Economic Theory, Vol. 3, p. 373 - 413.

[4] Modigliani, Franco and Richard Brumberg. 1954. “Utility Analysis and the Consumption Function:
An Interpretation of Cross-Section Data,” in K.K. Kurihara (ed.), Post Keynesian Economics. New
Brunswick.

11
 Money Notes

Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
Monetary theory and macroeconomics have long been intertwined. Many of the classical economists
thought that money played a significant role in economic fluctuations. Many economists in the era
immediately before Keynes also placed monetary policy, if not the money supply itself at the center of
their theories of economic fluctuations as well (see Laidler 1999, Ch. 2 - 5). Money also played an
important role in Keynes’s General Theory.1 Nonetheless, in the aftermath of Keynes, participants in
the so-called Keynesian Revolution largely downplayed the role of money in macroeconomic analysis.2
Interest in the role of money and business cycles as well as the quantity theory of money was revived
by Monetarists, such as Milton Friedman, Anna Schwartz, Karl Brunner, Allan Meltzer, and others.
These economists were also particularly interested in the monetary transmission process, or the way in
which monetary policy affects economic activity. In more recent years, a project by Michael Woodford
to imagine monetary policy in a hypothetical world without money has led to a full-fledged field of
monetary theory and policy analysis that ignores money completely.
Against this backdrop, what we are interested in asking: Is money important for macroeconomics? If
so, why? However, to answer these questions, we need to know several things. First, any analysis of money
requires a model, but we need to ensure that our model takes money seriously. To do so, we need to
understand why money exists and how money emerged. Second, whether monetary factors are important
for macroeconomic outcomes might also depend on the monetary regime. Such regimes have changed
significantly over time. Some of these changes were market based changes, but many were imposed by
states. In these notes, I will present a state-based theory of the evolution of monetary institutions. Third,
what sorts of claims have people made about monetary factors and the macroeconomy and how can
we examine and test these ideas? Fourth, we need to ensure that we are measuring money properly.
Otherwise, how reliable can our analysis of money really be? Fifth, conventional Walrasian monetary
models have a hard time incorporating money and explicit exchange. Thus, I present two types of models
(transaction cost models and search models) that are not only capable of providing both a micro-founded
model of money, but also capable for providing basic insight into monetary policy analysis. Nonetheless,
while these simple models can allow us to draw policy implications, these policy implications are based
on one and only one crucial friction in the economy: the lack of interest on currency. Thus, we must
ask ourselves how policy implications might change if we incorporate additional frictions into the model.
While we speculate on this last point, we will explore it formally in future lectures.
1
It is, after all, titled The General Theory of Employment, Interest, and Money!!
2
It is important to note that there were significant differences between Keynes and the Keynesians. This is far beyond the
scope of this course, but the interested reader is referred to Leijonhufvud (1968) and Davidson (2007, Ch. 12).

1
2 Why Money?
To understand money’s role in macroeconomics, we first need to understand the role that money plays
in the economy. Anthropologists have argued that not only did credit exist before money, but that early
societies were likely pure credit economies. So any discussion of the role of money must be able to
explain both why money exists and why it, to some degree, replaced credit. We will do this will a really
simple 3-person, infinite horizon model that cycles through 3 periods. This model is a variation of the
Kiyotaki and Moore’s (2002) model in “Evil is the Root of All Money” (who themselves borrowed the story
from the classical Swedish economist Knut Wicksell). That title is not a typo and it is not backwards.
You will understand why once we have worked through this simple model.
Consider an economy that consists of 3 people: Adam, Betty, and Chris. Adam produces applesauce
in period 1, but will not eat it. He does, however like to eat bread. Betty bakes bread in period 2,
but prefers a gluten free diet. She prefers to eat cheese. Chris produces cheese in period 3, but he’s
lactose intolerant so he would prefer not to eat cheese. He would prefer to eat applesauce. So we have
a 3 person economy in which the people have heterogeneous preferences and heterogeneous production
opportunities. Furthermore, since each person is producing food, let’s imagine that the periods are
sufficiently long that the food cannot be stored across periods. Finally, it is important to mention that if
we think about the preferences and the production capabilities of the people who populate our model, no
two individuals will ever be able to meet and engage in mutually beneficial exchange (there’s an absence
of double coincidences of wants, as economists like to say).
Given the absence of double coincidence of wants between any combination of two individuals, how
can trade ever take place? One way that trade could take place is if each of the participants are willing
to extent credit to one another. For example, in period 1, Chris could offer Adam an IOU in exchange
for applesauce. But would Adam be willing to accept the IOU? Well, consider that Adam wants to go to
Betty in period 2 and purchase some bread. Since he doesn’t produce anything that she would like to
consume, his only option would be to offer her an IOU. But would Betty accept an IOU? There is reason
to believe that she would if Adam transferred to IOU from Chris to her. For example, suppose that Adam
accepts the IOU from Chris. When he meets with Betty in period 2, he might say to her “Chris has given
me this IOU, but I don’t like cheese. I know that you like cheese, so I’d be willing to cash in my IOU
from Chris next period and give the cheese to you then if you give me the bread now.” Supposing that
everyone is willing to accept this arrangement, this is a pure credit economy. Trade is essentially being
traded for promises.
While it is possible that we end up in this pure credit economy, it is important to ask: under what
conditions would this arrangement hold up? The arrangement would work so long as everyone expects
Chris to produce in period 3. Since Chris consumes in period 1 and produces in period 3 and the world
only lasts for 3 periods, what incentive does Chris have to produce? While he might feel some moral
obligation, Chris does not have an economic incentive to produce. So if Betty does not believe that Chris
will produce in period 3, she will not be willing to accept an IOU from Adam in period 2. If Betty is
not going to accept his IOU, then Adam does not have an incentive to produce in period 1 for Chris.
The question is under what conditions would Chris renege on his promise. Consider two scenarios. In
the first scenario, Adam, Betty, and Chris all know one another and live in the same community. In the
second scenario, Adam, Betty, and Chris do not know one another and only encounter one another to
trade – as far as they are concerned the trade is anonymous.
In the first scenario, Chris has an incentive to keep his promise. If he does not keep his promise, then
Betty and Adam can punish him by excluding him from all future trade. In the second scenario, Chris
has no economic incentive to trade because trade is anonymous and there is no mechanism through
which he can be punished in the future.
Thus, in the first scenario, we have a pure credit economy in which trade is conducted through a

2
series of promises each cycle. In the second scenario, we have a world of autarky (and harsh dietary
restrictions). Nonetheless, the second scenario does not have to result in autarky. Suppose we introduce
money into this scenario. Chris can now present Adam with money in exchange for applesauce. Adam
will be willing to accept money if he believes that Betty will accept money. Furthermore, Betty will accept
money if she thinks that Chris willing be willing to accept money. As long as each of these individuals
thinks that they can use money for trade in the future, they will use it.
This outlines the central role for money in the economy. Money allows trade to take place that would
not take place otherwise. In our scenario, individuals would use credit if it was feasible. However, when
credit is infeasible, individuals use money instead. Money is therefore a substitute for credit. Money acts
as a record-keeping device demonstrating that trade has taken place (Ostroy 1973).
This really simple model explains why the earliest economies to trade used credit. In the earliest
economies, trade was local. Exchange was often between people who knew one another. These repeated
interactions meant that trade was not anonymous and that one’s reputation mattered. Someone who
valued trading with other people would not want to renege on a promise because others in the community
would find out and potentially punish them by excluding them from trade or, at the very least, refusing
to accept promises from them in the future.

3 A Search Theoretic Approach

To demonstrate this idea a bit more formally, we will consider the model of Kiyotaki and Wright (1993).
The economy consists of a continuum of agents with unit mass interacting in continuous time. Goods
and money are indivisible. There is one good and it can be exchange for one unit of money (say, one
“dollar”). This money is fiat money. Since money is indivisible, we will assume that only some fraction of
people in this economy are holding money (and they’re only holding one unit). We will assume that the
only people who produce goods are those who are not currently holding money. We will hereafter refer
to those who are not holding money as “producers.” We will assume that each individual in the economy
is endowed with a production technology to produce only one of n possible goods. The goods are not
storable over time.
Trade occurs through a random matching process. Suppose that the probability of being matched
with someone holding money is α and the probability of being matched with someone who isn’t holding
money be (1 − α). Once matched, the probability that an individual wants to consume the particular
good produced by their match is β. When trade occurs in a match, the producer incurs a cost C and the
consumer gets a benefit U .
Let V0 denote the value associated with someone who is not holding money and Vm denote the value
associated with someone who is holding money. It follows that the Bellman equations can be written,
respectively, as
rV0 = (1 − α)β 2 (U − C) + max Π(π)αβ(Vm − V0 − C) (1)
Π∈[0,1]

rVm = (1 − α)πβ(U + V0 − Vm ) − δm (2)

where π is the fraction of people willing to accept money, Π(π) is the best response probability that a
producer will accept money given the likelihood that others accept money, and δm is the storage cost
of holding money. Note that there are two opportunities for trade for producers. A producer can be
matched with someone holding money. In this scenario, the consumer will want to produce the good of
the producer with probability β. The producer then has to decide whether or not to accept money given
that the benefit the producer receives is Vm − V0 − C. It is also possible that the producer is matched
with another producer. In this case barter is possible if there is a double coincidence of wants. This
occurs with probability β 2 and the producer gets a benefit U − C.

3
 We will confine our analysis to a symmetric equilibrium in which Π(π) = π. Consider 3 different
possibilities. In the first case, the net benefit from trading with someone holding money is negative,
Vm − V0 − C < 0. In this case, no producer would be willing to accept money and therefore π = 0. In
the second case, the net benefit from trading with someone holding money is positive, Vm − V0 − C > 0.
In this case, all producers would be willing to hold money and π = 1. Finally, in the third case,
Vm − V0 − C = 0 and therefore producers are indifferent between producing and holding money. In this
case some fraction of producers will accept money and some fraction will not. For example, note that if
Vm − V0 − C = 0, then rVm − rV0 − rC = 0 and this implies that

(1 − α)πβ(U + V0 − Vm ) − δm − (1 − α)β 2 (U − C) − παβ(Vm − V0 − C) − rC = 0

Recall that Vm − V0 − C = 0 and therefore V0 − Vm = −C, this expression can be simplified to

(πβ − β 2 )(1 − α)(U − C) − δm − rC = 0

We can solve this expression for π. Thus, a monetary equilibrium will exist if π = 1 or if π = π̂ where

rC + δm
π̂ = +β (3)
β(1 − α)(U − C)

Let’s think about the implications here. In some sense, what we have is a model with the same basic
features as our 3-person economy above. In this search model, like our 3-person economy, the decision
about whether to accept money depends on whether one thinks that others will be willing to accept
money in future transactions. According to our framework, if one expects a fraction of people π̂ to
accept money, then individuals are indifferent between accepting money and not and the result is that
an exact fraction, π̂, will accept money. If, however, individuals expect a fraction that is smaller than
π̂ to accept money, then Vm − V0 − C < 0 and nobody will be willing to accept money and π = 0 in
equilibrium. If individuals expect a fraction of people larger than π̂ to accept money, then Vm −V0 −C > 0
and everyone is better off by accepting money. It follows that everyone will accept money in that case.
This shows that there is a network effect that explains why money exists in equilibrium. In other words,
people are willing to accept money in equilibrium because they anticipate that a certain proportion of
other people will be willing to accept money from them in the future.
It is important to note that our model implies that this is true even if money is intrinsically useless.
This is an important point because it suggests that fiat money can be valued in equilibrium and that
money does not have to be a commodity. Nonetheless, if we look at the emergence and evolution of
money, only commodity money can be said to have “passed the market test.” Non-commodity monies,
in other words, did not spontaneously emerge through trade or through some competitive process. In
fact, non-commodity monies have only been introduced through government fiat or some other form of
intervention. As a result, we need to understand the evolution of money and monetary institutions.

4 On the Origins of Money

The previous two sections outlined the role that money serves in the economy. Money exists when credit
is not feasible. Understanding why money exists, however, does not tell us how money emerged or what
sorts of things might be used as money. In our simple model, we simply introduced money and explained
why it solve the problem of the infeasibility of credit. But where did money come from?
Both Adam Smith (1776) and Carl Menger (1892) provide us with speculative accounts of how money
emerged and evolved over time. Adam Smith argues that money emerged in conjunction with the divi-
sion of labor. As labor becomes more specialized, consumption became detached from production. This

4
requires trade, which is difficult without money (Adam Smith seems to view barter as the alternative to
money rather than credit, but his story works nonetheless). The earliest forms of money were commodi-
ties. Adam Smith notes that cattle served as a medium of exchange during the time of Homer. Salt was
used as money in what we now call Ethiopia. In India, shells served as money at one time. Tobacco was
used in Virginia. Nonetheless, as Adam Smith points out in most plays metals ultimately replaced these
other commodities as money. The convergence to metals created one problem. If metals were exchanged
for goods, this required a particular quantity of the metal to be exchanged for a particular quantity of
the good. The quantity of metal was often specified by its weight. However, metals have differing degrees
of fineness. Using metals as a medium of exchange perhaps reduced transaction costs in comparison to
cattle or shells or tobacco, but transaction costs still remained since exchange required examining the
weight and fineness of the metal. As a result, coins were developed with the weight and fineness stamped
onto the metal.
Metallic coins are not without inconvenience either.3 For example, coins were often odd mixtures
of precious metals. Differences in accounting units and particular coins unique to a local area create
substantial transaction costs. This leads to the first type of banking, which were the so-called money
changers. Money transfer services were around at least since medieval Europe. These money changers
would often trade local coins for foreign coins and accept deposits for safekeeping. The acceptance of
deposits led to another innovation, which was the ability of individuals to trade without the physical
exchange of coins. Since money changers kept a ledger, trades could take place through transferring
balances on the ledger.
Nonetheless, modern banks serve two essential roles, they issue deposits that can be used as media of
exchange and they serve as intermediaries between borrowers and lenders. Money changers only served
the first purpose. Selgin and White (1987) argue that this aspect of banking emerged in the 17th century in
England. During this time, goldsmiths started accepting deposits for safekeeping and allow ledger-based
transfers of gold. Before long, there is an innovation in this gold warehousing when it because recognized
that (a) money is fungible, and (b) holding fractional reserves is permitted by the law of large numbers. In
other words, these warehouses came to realize that they could lend some of this money out to borrowers
without any threat to the warehouse’s ability to meet the withdrawal requests. Since daily withdrawals
and deposits tend to cancel out, or at the very least, net withdrawals on any given day are but a small
fraction of total deposits, the warehouses could manage a fractional reserve system. These warehouse
banks therefore became intermediaries before the Glorious Revolution in England.

5 The Evolution of Monetary Institutions: A State-Based Approach

In the previous section, we outlined how money evolved and what sorts of things might serve as money
in this evolutionary process. Historically, things like gold and silver were money and private banks issued
banknotes convertible into these commodities. Today, we live in a world in which central banks have
a monopoly over the issuance of currency. In addition, the currency in circulation is not convertible
into any given commodity. Whereas in the past, the word “dollar” was literally defined in terms of gold
and silver, today the word “dollar” is just green pieces of paper with different denominations issued by
a single bank. So your $20 bill is no longer redeemable for roughly an ounce of gold, but instead is
redeemable for a $10 bill and two $5 bills – or some other combination of smaller denominations. This
begs two questions:

1. Why do we have fiat money? (Or, how did we get here?)

2. Why are people willing to accept fiat money?

3
What follows draws on Selgin and White (1987) and Glasner (1989a).

5
The second question is in some sense easier to answer. In our 3-person economy of section 2, we showed
that the money serves as a substitution for credit. In section 3, we showed that when trade is anonymous
the acceptance of money is driven by network effects, even if its intrinsically worthless. In theory, there
is no reason why this thing that people pass around as a medium of exchange has to have some sort
of intrinsic value separate from its exchange value. In practice, what actually emerged as money were
things that have some intrinsic value besides their use in exchange. What this suggests is that we might
want to consider how monetary institutions evolved and why they evolved the way that they did.

5.1 The Historical Evolution of Money

Our discussion of money has emphasized its role as a medium of exchange. However, if money is a
medium of exchange, then it is somewhat natural for it to also be a medium of account. In other words,
if we are exchanging money for goods and goods for money, then it would make sense to post price in
terms of money. If money is a precious metal, then a natural way to post prices is in terms of the weight
of the metal. Weighing and measuring money for each transaction adds to transaction costs. So, as we
noted above, metallic money gives rise to coinage. When someone mints a coin, the coin can then be
stamped with its weight, which reduces transaction costs. According to Arthur Burns (1927), the earliest
known coinage occurred in the seventh century B.C. in Lydia. However, as we learn from David Glasner
(1989a), this is also the first place in which the state became involved in coinage. Just a century later,
the Greek city-states were involved in minting coins as well. In addition, the Greek city-states ensured
that they had a virtual monopoly over minting by requiring that taxes were paid in coins minted by the
city-state. This monopoly over minting and the requirement that taxes be paid with these coins not only
provided a revenue source during peacetime, but also represented a way that the city-states could raise
revenue quickly to finance defense.
But remember that money is not limited to coins. Banks emerged and issued their own banknotes.
However, if we look back at history, the evolution of money and banking has the same general theme that
emphasized with coinage. Governments know that a monopoly over money provides important benefits
in the form of seigniorage. Government control over money can therefore be an important source of
revenue. Private note issuance is a threat to this revenue. As a result, we would expect that the history of
banking regulation, especially with regards to note issuance, can be explained by the government’s desire
to maintain seigniorage.
Shortly after the Glorious Revolution in England, the Bank of England was chartered by Parliament
on the condition “that the stockholders buy £1,200,000 of government bonds yielding 8 percent annual
interest” (Glasner 1989a, p. 34). The implication being that the government needed to raise revenue and
the charter of the Bank of England gave the British Crown access to credit that would not have otherwise
been available. Eventually, the British gave the Bank of England a monopoly over currency issuance
in 1844 with the Bank Charter Act. During the Civil War, the United States chartered national banks,
which were required to hold government debt to back their note issuance. These nationally chartered
banks competed alongside state-chartered banks that, as part of the corresponding bank regulation, had
their banknotes taxed. At the same time, the U.S. government issued so-called Greenbacks, which were
inconvertible pieces of paper to finance war-time expenditures. It was only after the unification of the
German states that Bismarck introduced a central bank with a monopoly of currency issuance. The Bank
of Italy was given a monopoly over currency issuance by Mussolini in 1926.
What this general pattern highlights is that, while money emerged spontaneously to meet the needs
of trade, it is the government’s desire for revenue that ultimately shaped and molded the monetary
institutions that we have today. The government’s involvement in minting was a way to raise revenue
before effective taxation was possible. The creation of central banks was to provide a source of revenue
and emerging financing for the government. Bank regulation, such as the requirement that banks back

6
their notes with government debt and the taxation of private banknotes out of existence, while inefficient
in a purely economic sense, was a way to limit competition in banking to generate revenue for the
government. The same can be said about the shift to monopolies over currency. No monopoly over
currency ever emerged in the course of normal market activity. Currency monopolies have only been
granted by governments, not economies of scale of some other mechanism. Furthermore, inconvertible
paper money did not exist except by government fiat. But before we conclude that this was purely a
nefarious development on the part of the state, we might stop to consider whether this process actually
had benefits.

5.2 Did This Evolution Serve an Important Purpose?

The various government regulations of banking and note issuance are deliberately self-serving. But the
fact that these regulations were self-serving does not necessarily mean that they did not provide some
external benefits. For example, there is a literature on what is known as state capacity that emphasizes
the role of the development of the state in conjunction with economic development.4 The idea is that the
ability of the state to effectively tax to generate revenue and the ability of the state to provide defense
against both internal and external predation, created the conditions for economic development. The
various regulations of money and banking by the state can be thought of in this context. For example,
a monopoly over coinage, note issuance, and even the creation of a central bank, provided important
sources of open-ended government finance if the state found itself at war.
Thus, early monopolies over coinage created the ability for states to raise revenue when taxation was
difficult and to raise funds to pay for defense. The charter of the Bank of England provided an important
source of credit for British should they end up needing funds for national defense. As Thompson and
Hickson (2001) argue, the gold standard itself was an important institution for providing defense. They
argue that the gold standard created the expectation of price level stability. When emergencies presented
themselves, like a a military threat from a foreign power, the government would need to raise revenue.
As a result, they would suspend the convertibility of currency into gold and print money to finance the
war. However, once the war was over, they would restore convertibility into gold. The commitment to
returning to the gold standard was important. Without this expectation, printing money would result in
a proportional increase in the price level. Thus, the commitment to restore the gold standard caused the
price level to increase less than proportionately to the increase in the supply of banknotes, which meant
that printing money produced real purchasing power for the government. Furthermore, they argue that
after the creation of nuclear weapons, the need for emergency financing for long conflicts diminished,
which enabled the transition from a gold standard to the fiat regimes that exist today.
Economists would do well to keep this understanding in the back of their head when they are
evaluating historical experience with bank regulation and currency monopolization. Sure, from a pure
economic efficiency standpoint a free banking system would work better than a government monopoly,
but a holistic theory of economic development suggests that the economic inefficiency of monopoly might
have actually been institutionally efficient from the perspective of fostering the development of the state,
the provision of defense, and the wealth creation that subsequently resulted.
As a way of tying all of this stuff together, I would like to emphasize that the reason that these notes
are so long and cover as much history as they do is that money is intertwined with just about everything
that is important in macroeconomics. Money plays an important role at the microeconomic level of
trade. At the aggregate level, the way in which the money supply expands and contracts (or fails to, as
it were) have important implications for macroeconomic outcomes and business cycle research. Finally,
the structure of monetary institutions, the way that they have evolved, and the way they exist at present
are part of a path dependent structure that is a story of both the development of the state and economic
4
See the survey by Johnson and Koyama (2017).

7
development. Thus, before we go about assuming that we can change monetary policy and/or monetary
institutions, it is important to consider the path dependent nature of these institutions.

5.3 Bretton Woods, the Collapse, and Fiat Money. Why Does Anyone Accept This
Stuff?
The history that I presented provides a sort of revenue-based/defense-based explanation for the evolution
of monetary institutions. However, that history really only takes us up through the period of time that
corresponds to the gold standard. In planning the post World War II monetary system, countries were
anxious to avoid the sort of deflation experienced during the interwar period. Nonetheless, countries
still wanted to maintain some version of the gold standard and the opportunities for emergency finance
that went along with it. Thus, in theory, the countries wanted a system in which central bank could
promise a credible commitment to price stability without the severe threat of deflation that went along
with resumption after emergency suspensions (Thompson, 1997).
The Bretton Woods agreement was an attempt to accomplish this goal. The Bretton Woods system
was one in which countries agreed to fix their exchange rate relative to the dollar. However, to ensure
price level stability, the U.S. agreed to maintain a constant price of gold. Note the difference here between
the Bretton Woods system and the gold standard. Under the gold standard, currency was redeemable
for gold on demand. Under the Bretton Woods agreement, there was no such option of convertibility.
Instead, this was a gold exchange standard. Central banks would hold dollars in reserve to maintain
their fixed exchange rate. If these central banks ever became concerned about the stability of the dollar,
they could always exchange the dollars they held in reserve for gold.
The Bretton Woods system worked well until the 1960s when the United States created too much
inflation for the system to survive. Niehans (1979) provides an excellent overview of the system and the
theoretical foundations thereof. He argues that in order for the system to work, the United States had to
make sure that dollars and gold remained perfect substitutes from the perspective of central banks. The
inflationary policy of the Federal Reserve in the 1960s reduced the demand for dollars. As a result, the
U.S. saw its gold reserves decline. In order to maintain the system, the United States would have had to
devalue the dollar or engage in contractionary monetary policy. The U.S. was unwilling to do this and
President Nixon closed the gold window. The collapse of the Bretton Woods system brought about the
beginning of truly fiat money.5
Thus, what we need is a theory that is capable of including fiat money that is held in equilibrium and
the role for monetary policy in such a system. Before we get there, however, we want to take a detour to
discuss the importance that money has played in macroeconomic thought. This detour might help us to
reach a framework capable of taking both money and macroeconomics seriously – or at the very least,
convince you it belongs in macroeconomic models.

6 Money and Macroeconomics

Money has often played a significant role in the discussion of economic fluctuations. As the previous
section detailed, however, there have been significant changes in monetary institutions over time. In this
section I will discuss the role of money in macroeconomic analysis and how the role of money changes
when the monetary regime changes.
5
Thompson (1997) argues that the decision to close the gold window can be understood within the context of the need for
emergency defense discussed above. Specifically, given the nuclear power structure, there was now no longer any need for
emergency finance to the same degree as in the past. As a result, the shift to fiat money was politically feasible.

8
6.1 The Gold Standard, Classical Monetary Theory, the Quantity Theory of Money,
and Implications of Different Regimes
So many of the debates in macroeconomics that involve money have to do with the gold standard. In
fact, a number of debates in macroeconomic thought can be understood if one understands (a) how the
gold standard worked, and (b) why one has to use a different monetary theory depending on whether the
gold standard was operational. Many historical debates can be understood as the failure by one side or
the other to recognize that whether a particular theory was applicable or depended on whether money
was convertible into gold. We will discuss this further in the next section, but before we can, we need to
first understand how the gold standard worked. To get this understanding, we will rely on the work of
McCloskey and Zecher (1976).
Let’s start with an idea that is wrong, so we can make sure to avoid it in all future endeavors. This is
the idea of the price-specie-flow mechanism introduced by David Hume. According to Hume, an increase
in the money supply would cause domestic prices to increase. This would lead to a decline in exports
and a gold outflow. The money supply would then have to fall in order to ensure the convertibility into
gold. The price level would then return to its original level. The basic idea here is that (1) deviations in
price levels across countries are caused by changes in the money supply, and (2) gold flows necessitate a
reversal in the money supply. This view gives rise to a belief that participants in the gold standard must
follow the “rules of the game” in which they adjust the money supply in accordance with gold flows.
The problem with the price-specie-flow mechanism is that the implications are not supported by the
data. For example, McCloskey and Zecher (1976) point out that gold flows are much too small for this
idea to be correct. Second, and most importantly, the price-specie-flow mechanism suggests that when
there are deviations between the price levels of gold standard countries, such deviations will be corrected
by gold flows in conjunction with the rules of the game. However, when one looks at the data, the
evidence suggests that price levels throughout the world are positively correlated rather than negative
correlated even in the short run. Over a long time horizon, we would expect price levels to be positively
correlated. In the short run, we should expected to observe an equilibrium correction mechanism. This
is not evident in the data.
So if the price-specie-flow mechanism does not describe how the gold standard worked, then how
did it work? Let’s think about price theory. When we plot supply and demand curves, we are looking at
what happens to supply and demand given a change in the relative price of a good. In other words, if we
are looking at the market for one good, then we are looking at the price of that good relative to all other
goods. Thus, when we think about the market for gold, we are thinking about the price of gold relative
to all other goods. Fluctuations in supply and demand for gold affect the relative price of gold. Since
gold is traded internationally, arbitrage will ensure that the law of one price will hold. Now consider that
under a gold standard, banknotes were redeemable at a fixed rate with gold. In fact, monetary units such
as the dollar were literally defined as a particular quantity of gold. Put differently, the price of gold in
terms of dollars (or any other monetary unit) was fixed. Thus, by ensuring that the law of one price held
for gold, fluctuations in the relative price of gold were reflected in changes in the price level. According
to this view, fluctuations in the gold market caused the price level in gold standard countries to move
positively with one another. As the price level rises, this increases the demand for money and therefore
banks (and/or central banks) have an incentive to expand note issuance.

6.2 Classical Monetary Theory, the Quantity Theory of Money, and Implications of
Different Regimes
Given that we are not only the gold standard, Monetary theories of economic fluctuations can broadly
be classified into two categories: classical monetary theory and the quantity theory of money. Which

9
theory to use to analyze particular monetary questions depends on the monetary regime. Numerous
historical macroeconomic debates can largely be classified as one group applying the wrong theory to
the then-current monetary institutions (Glasner, 1989b). For example, under a gold standard the quantity
theory is only appropriate if the price-specie-flow mechanism is operational. As we just outlined, the
evidence does not support this idea. Rather, it is the price of gold that pins down the price level and
the money supply adjusts to corresponding changes in money demand. In the absence of convertibility,
the price of gold no longer pins down the price level. As a result, the quantity theory is the appropriate
framework for thinking about fluctuations in nominal income and the price level.
We will discuss the quantity theory of money at more length when we discuss Friedman and Schwartz
below. For now, I want to focus on the classical theory. I want to do so because in the subsections that
follow, we will discuss explanations of the Great Depression, the stagflation of the 1970s, and why it is
important to apply the correct theory.
It is hard to articulate the classical theory. However, I want to follow David Glasner (1985) and
take a roundabout way toward discussing the classical theory. What I mean is that Earl Thompson
(1974) developed his theory of money and income and found that many of the conclusions of classical
economists were correct when viewed in the context of his theory, despite the fact that modern economists
would criticize classical economists for making such claims. Thus, in a roundabout way, Earl Thompson’s
theory is the classical theory of money and it is easier to understand. So we will treat Thompson’s model
as the classical theory.
Thompson begins with an analysis of orthodox value theory. He argues that within a standard
competitive model, for money to exist, it must produce the competitive equilibrium even though money
is used to trade. In other words, traditional explanations that rely on various transaction costs and other
frictions must depend on such costs and frictions existing in disequilibrium – otherwise the competitive
results would not hold. So how can money exist within the context of the standard competitive model?
Suppose that there is some fraction of individuals in the economy that are easily recognizable,
endowed with a technology to costlessly screen and monitor potential borrowers, and have a public
reputation because of their recognizability. We will call these people banks. All other individuals in
the economy are not recognizable and possess no such technology. These banks will have an incentive
to be producers of money. Why? Consider our 3-person economy of section 2 or our discussion of
Kiyotaki and Wright in section 3. In these frameworks money is necessary to facilitate trade when trade
is anonymous. However, the willingness of individuals to accept money, however, is dependent upon a
network effect (i.e., people accept money if they expect a sufficient number of others to do the same).
In those models, however, we never made mention of where money came from. We simply asked if
people would be willing to accept money given that money existed. Thompson argues that money will
emerge competitively because banks will have an incentive to produce it. Banks, since they are universally
recognizable can always be excluded from trade if they misbehave. Banks therefore have an incentive
to develop a reputation. A bank could then produce money that people would be willing to accept on
the basis of that reputation. A number of economists have argued that competitive note issuance poses
problems. The argument is that because value theory implies that price will be equal to marginal cost
and because the marginal cost of producing money (banknotes) is arbitrarily close to zero then the price
is zero. Money with a price of zero has no value and will not circulate. Put differently, the argument is
that competition in the production of money will lead to excess money printing which ultimately results
in money being worthless. However, this ignores the crucial role of property rights implicitly assumed to
be in the background of the standard competitive model. With private property rights, a bank that over-
issues and reduces the value of all other banknotes in the process would be infringing on the property
rights of other banks. Thus, banks will have the right to produce banknotes that are unique in some way.
In addition, banks need to develop a reputation to ensure that their notes will continue to be accepted.
The bank can then buy interest-bearing assets (i.e., loans or real assets) from non-banks in exchange

10
for banknotes and the non-banks can use these banknotes. But what can the bank do to ensure that
their particular notes are used? One way to do so is to promise to buy back the notes in exchange for
a commodity at a fixed price in the future. Banks that initially do this will find that they attract more
people to use their notes than those of their competitors. Competition will therefore result in all banks
offering banknotes convertible into this commodity.
In this economy, the supply and demand for the commodity for which money is redeemable will
determine the real price of this commodity. Since the nominal price of the commodity is fixed, the
price level is pinned down by the supply and demand for the commodity. Banks are driven by profit
motivations to produce whatever quantity of money that the public wants to hold and therefore the supply
of money is perfectly elastic at a given price level.
So, you might be asking yourself at this point, what does this have to do with the macroeconomy?
Well, Thompson’s theory has several implications. First, since the money supply is competitively supplied,
any change in the demand for money will have no effect on economic activity. Instead, banks will be
motivated by profit to adjust the money supply in accordance with money demand. Second, changes in
the money supply will have no effect on economic activity. If banks issue too many banknotes, since
money is convertible into a commodity, the law of reflux will kick in. Individuals will show up at banks to
redeem their notes. This drain on reserves would force banks to restrict their note issuance. Third, there
is no such thing as a real balance effect. Fluctuations in the commodity market will cause fluctuations
in the price level. If the price level rises (falls), real money balances decline (rise). Individuals will want
to increase (decrease) their nominal money holdings and the competitive banking will accommodate this
change in demand. Fourth, consider a decline in the marginal product of capital. This causes the rental
rate on capital to decline. With a competitive money supply, this reduces the rate of return to bankers
and therefore bankers pay a lower rate of return on money. Fifth, what causes fluctuations in this model
is a change in the relative price of the good for which money can be redeemed. Changes in the relative
price of gold, because the nominal gold price is fixed, must come in the form of changes in the overall
price level. If wages are sticky, this implies that fluctuations in output prices will lead to changes in real
economic activity.
It follows that in the classical theory, by assuming a competitive money supply, economic fluctuations
are caused by changes in the relative price of gold. This theory is also consistent with our discussion
of how the gold standard worked, in which the supply of money fluctuates in accordance with money
demand and that the price level is pinned down by the price of gold.6 Given this discussion, we can now
discuss the gold standard in the context of this classical theory.

6.3 The Thompson-Glasner-Sumner Theory of the Great Depression

Earl Thompson’s work on monetary theory suggests that changes in the relative price of gold are the
source of significant fluctuations in economic activity under the gold standard and that, in fact, this is
what caused the Great Depression:
There can be little doubt that classical monetary analysts, who were also inveterate policy
advisors, put little or no weight on the recessionary significance of a rise in the relative
demand for the asset backing money, i.e., gold. The policy neglect of these shifts permitted
a series of sharp recessions caused by sudden increases in the demand for gold through the
history of Europe and the United States. . . The last such recession was the Great Depression,
which saw a five-year increase in the real price of gold (1929 - 1934) whose magnitude was
6
This idea was clearly recognized by Adam Smith when he wrote the Wealth of Nations. Smith was familiar with Hume’s
price-specie-flow mechanism, but rejected this view. Smith’s views would later be adopted (incorrectly) by the Anti-Bullionists
and (correctly) adopted by the Banking School in the famous 19th-century monetary debates in the U.K. See Humphrey (1981),
Laidler (1981), and Glasner (1989a) on understanding Smith’s monetary theory.

11
 unprecedented in recorded history, an increase resulting from the return to the gold standard
from 1924 to 1928 (Thompson 1974, p. 449).

Scott Sumner (2015) presents a theory of the Great Depression consistent with Thompson’s account. He
uses a rather simple way of thinking about the gold market to account for economic fluctuations. For
example, if economic fluctuations are driven by changes in the relative price of gold, then what we need
is to understand what affects the relative price of gold. Sumner starts with the following equilibrium
condition:
Pg gs
= P md (4)
r
where Pg is the price of gold, gs is the stock of gold, r is the gold reserve ratio, P is the world price
level, and md is the demand for currency. Here, Pg gs is the nominal value of the stock of gold. Given
the gold reserve ratio, r, this implies that the left-hand side is the nominal supply of currency. The
right-hand side measures the nominal demand for currency. Market-clearing implies that the two sides
must be equal. According to Thompson and Sumner, what we are really interested in is what happens
to the relative price of gold since it is the relative price of gold that has a significant effect on economic
activity. Solving this equation for the relative price of gold yields:

Pg rmd
= (5)
P gs
Thus, on the right-hand side, the numerator represents the demand for monetary gold and the denom-
inator represents the supply of monetary gold. When the demand for monetary gold rise, the relative
price of gold rises. When the supply of monetary gold rises, the relative price of gold declines. However,
remember that under a gold standard the price of gold is fixed. So in order for these relative price
changes to take place, the price level must adjust. Since the price of gold is fixed, we can also normalize
the price to 1 and write our equation in terms of the price level:
gs
P = (6)
rmd
Sumner writes this equation in log-differences to account for the changes in the (world) price level over
time:
∆ ln P = ∆ ln gs + ∆ ln(1/r) + ∆ ln(1/md ) (7)
This equation is simply an account device to determine which of these factors were most important in
explaining the movements in the price level. I re-print part of his Table 13.3 as Table 1.

Table 1: The Impact of the Supply and Demand for Gold on the World Price Level
Time Dec. 1926 - Dec. 1932
∆ ln(1/r) -21.86
∆ ln(1/md ) -40.80
∆ ln gs 25.61
∆ ln P -37.06

It is important to point out that this table addresses the world price level because the price level
in each country should move in the same direction in all countries on the gold standard. The only
differences in the movements in the price levels across countries should be in non-tradeables, since there
are no opportunities for international arbitrage. As shown in the table, the price level declined by 37%
during this six year period. Sumner shows that from Dec. 1926 to Oct. 1929, the price level only declined

12
by about 3%, which means that the vast majority of this decline occurred from Oct. 1929 - Dec. 1932.
The accounting framework provided by Sumner suggests that the demand for gold increased by over 60%
during this time frame whereas the supply of monetary gold only increased by 26%. The fact that the
demand for gold was increasing by the more than supply suggests that the price of gold had to increase.
Since the price of gold is fixed under the gold standard, this meant that the only way that the real price
of gold could increase is if all other prices declined.
Of course, this is just accounting. We need an economic explanation. In other words, the gold
market has to remain in equilibrium regardless of whether fluctuations in this market are the cause or the
effect of economic fluctuations. So how can we determine whether the effects were coming from the gold
market? Both Glasner (1989a) and Sumner (2015) provide evidence that the gold market is the source of
fluctuations. Glasner largely emphasizes changes in monetary policy whereas Sumner discusses the role
of both policy and private gold hoarding. As Sumner’s book details, while other theories of the Great
Depression can explain some of the observed fluctuations in economic activity and the price level, only
the gold market approach can explain all of the fluctuations.
Here is a sketch of the narrative. During World War I, the participants of the gold standard had
suspended the convertibility of currency into gold. Following the war, there was a desire to return
to the gold standard, but there was a question about how to go about such a return. For example,
there had been a significant inflation as a result of the suspension of convertibility. This meant that
restoring the previous parity to gold would require a significant deflation. To understand why, consider
that when convertibility is suspended, this reduces the demand for monetary gold from central banks.
When central banks suspend convertibility, the quantity theory kicks in and inflation ensues (the extent
to which inflation occurs depends on expectations about the resumption of convertibility, see Thompson
and Hickson 2001). In addition, this reduction in the demand for monetary gold causes inflation in those
countries that remained on the gold standard as well. When convertibility is restored, this increases the
demand for monetary gold. An increase in the demand for monetary gold on the part of central banks
means that the real price of gold must increase. One way to achieve this would be to raise the nominal
price of gold (devalue the currency). The alternative in which the currency is returned to pre-war parity
requires deflation. During the suspension central banks had significantly expanded the nominal value of
their liabilities. In order to demonstrate a commitment to redeem currency at the pre-war parity would
require individual central banks to accumulate gold reserves. However, as the gold market equilibrium
condition implies above, a higher nominal value of currency at the pre-war parity would require lower
gold reserve ratios. As Glasner (1989a, p. 114) mentions, the need for central banks to attenuate their
demand for gold to avoid a severe deflation seems to have been well known. Nonetheless, there was a
coordination problem. In order to have avoid a severe policy-induced deflation would require central
banks to limit their collective demand for gold. At the same time, each central bank had an incentive to
increase their gold reserves to signal their commitment to convertibility.
While no formal agreement was reached to limit central bank demand for gold, some steps were
taken to this end. For example, countries that resumed convertibility did not re-introduce gold coins.
This meant that currency was redeemable for gold bullion. The lack of gold coins in circulation meant a
lower demand for gold. In addition, the Federal Reserve, which had accumulated a substantial fraction
of the world’s gold, allowed gold outflows throughout the 1920s. This was important because it meant
that the increased demand for gold from other countries was at least being partially offset by a reduction
in the demand for gold in the United States. Ultimately, however, this was short-lived. In 1928, France
returned to the gold standard. However, the legislation that returned France to the gold standard required
the Bank of France to have a gold reserve ratio of 35% and prohibited France from acquiring additional
foreign exchange reserves. As a result, the Bank of France started accumulating gold reserves by selling
foreign exchange, which increased the demand for monetary gold. Whereas the Federal Reserve had
been reducing its demand for monetary gold, this policy was reversed in 1928.

13
 This brings us to the stock market crash of 1929. The most logical explanation of the crash is that of
Glasner (who credits Earl Thompson with this explanation). David Glasner argues that the resumption
of the gold standard created expectations of deflation, but that when the Federal Reserve remained
committed to allowing gold outflows during the 1920s, this created an expectation that things would be
better than previously thought. This change in expectations led to a stock market boom since investors
started raising their expectations of future earnings. However, in 1928 and 1929 the Federal Reserve
became concerned about the level of the stock market and stock market speculation. As a result, the
Federal Reserve raised the discount rate and increased its demand for gold. The increase in world
demand not only increased the real price of gold, but also signaled to market participants that the Fed
would prioritize domestic returns and that deflation was to come. As a result, the stock market crashed
and the high real price of gold depressed economic activity throughout the world. Subsequent bank
failures led to an increase in the demand for currency, which increased the demand for monetary gold
and therefore the price of gold. Furthermore, the bank failures and fears of currency devaluation led to
private gold hoarding in the United States, which further contributed to deflation.
Perhaps the strongest evidence in support of the Thompson-Glasner-Sumner theory is the fact that
countries that left the gold standard started their recoveries sooner. Sumner (2015) shows, for example,
that the decision of the United States to leave the gold standard led to a dramatic increase in industrial
production and a significant stock market boom. However, as Glasner (1989a), Cole and Ohanian (2004),
and Sumner (2015) argue, the Great Depression did not end for some time because of the New Deal
policies.

6.4 Friedman and Schwartz: A Monetary Theory of Nominal Income

6.4.1 An Overview
To say that Milton Friedman and Anna Schwartz’s Monetary History of the United States had a significant
effect on the profession would be a major understatement. The book, perhaps more than any other work,
marks the beginning of the so-called Monetarist Counterrevolution. Friedman and Schwartz also wrote
a summary of their work in a paper entitled “Money and Business Cycles,” which we will draw upon in
these notes (because I would like you to actually read it). As the title implies, Friedman and Schwartz
argue that monetary factors have played an important role in all of the major fluctuations in the United
States between 1870 and 1960.
In his review of the book, Allan Meltzer argued that the underlying framework for the analysis is the
quantity theory of nominal income. Friedman and Schwartz (1982) later endorsed that view. To get a
better understand of this idea, consider that the following modified equation of exchange:
 
k+1
BV = P Y
k + rr
where k is the currency-to-deposit ratio, rr is the reserve-to-deposit ratio, B is the monetary base, V
is the velocity of money, and P Y is nominal income. The term in parentheses is commonly referred
to as the money multiplier. Friedman and Schwartz argued that fluctuations in monetary factors, k, rr,
and B were the major determinants of fluctuations in nominal income. Furthermore, given informational
frictions that prevent immediate adjustment from one equilibrium to another in the immediate term,
these fluctuations in nominal income were divided between prices and real economic activity. To see that
this was at the core of their theory, consider the following explanations for business cycles from their
paper:
1875 - 78: Political pressure for resumption led to a decline in high-powered money, and the
banking crisis in 1873 and subsequent bank failures to a shift by the public from deposits to
currency and to a fall in the deposit-reserve ratio.

14
 1892 - 94: Agitation for silver and destabilizing movements in Treasury cash produced fears
of imminent abandonment of the gold standard by the United State and thereby an outflow
of capital which trenched on gold stocks. Those effects were intensified by the banking panic
of 1893, which produced a sharp decline, first in the deposit-currency ratio and then in the
deposit-reserve ratio.
1907 - 08: The banking panic of 1907 led to a sharp decline in the deposit-currency ratio and
a protective attempt by banks to raise their own reserve balances, and so to a subsequent fall
in the deposit-reserve ratio.
1920 - 21: Sharp rises in Federal Reserve discount rates in January 1920 and again in June
1920 produced, with some lag, a sharp contraction in Federal Reserve credit outstanding,
and thereby in high-powered money and the money stock.
1929 - 33: An initial mild decline in the money stock from 1929 to 1930, accompanying a
decline in Federal Reserve credit outstanding, was converted into a sharp decline by a wave
of bank failures beginning in late 1930. Those failures produced (1) widespread attempts
by the public to convert deposits into currency and hence a decline in the deposit-currency
ratio, and (2) a scramble for liquidity by the banks and hence a decline in the deposit-reserve
ratio. The decline in the money stock was intensified after September 1931 by deflationary
actions on the part of the Federal Reserve System, in response to England’s departure from
gold, which led to a still further bank failures and even sharper declines in the deposit ratios.
Yet the Federal Reserve at all times had power to prevent the decline in the money stock or
to increase it to any desired degree, by providing enough high-powered money to satisfy the
banks’ desire for liquidity, and almost surely without any serious threat to the gold standard.
1937 - 38: The doubling of legal reserve requirements in a series of steps, effective in 1936
and early 1937, accompanied by Treasury sterilization of gold purchases, led to a halt in the
growth of high-powered money and attempts by banks to restore their reserves in excess of
requirements. The declines in the money stock reflected largely the resultant decline in the
the deposit-reserve ratio. (Friedman and Schwartz 1963; reprinted in Friedman 1969 , p. 217
- 218).

In each case, they highlight monetary factors that played a significant role in the context of the equation
of exchange above. The book was particularly important when it came to the Great Depression. In the
immediate aftermath of the Great Depression and the consensus was that the Depression represented a
failure of the capitalist system. The Keynesian Revolution and the various offshoots thereof sought to
explain the failure of markets to “correct” and return to a full employment equilibrium. Friedman and
Schwartz presented a different view. They blamed the Federal Reserve for its poor policy. They argued
that the Federal Reserve failed to meet the increased demand for currency, failed to serve as lender of
last resort, and engaged in unnecessary tightenings in monetary conditions.

6.4.2 The Classical Theory, the Quantity Theory, and Monetary History
Recall from our discussion above that the quantity theory is not valid under a commodity standard. The
price of gold pins down the price level, not the money supply. Thus, one open question is how to square
the quantity theory explanations of Friedman and Schwartz given that banknotes were redeemable for
gold in the U.S. during virtually this entire period that they examined. One way to do this is to think
about the quantity theory in the way that Milton Friedman did. He stressed the economic fluctuations
were driven by deviations between the supply of and the demand for money. According to the classical
theory, which is consistent with a commodity standard, this should not be possible. The money supply

15
should adjust to changes in money demand and only changes in the relative price of gold should matter.
So how could changes in the money supply be important? I will present one possible hypothesis.
The classical theory assumes a competitive money supply. Prior to the creation of the Federal Reserve
in 1913, the U.S. did have a competitive money supply. However, the degree to which banks notes could
respond to changes in the demand for banknotes or substitution between deposits and banknotes was
limited by regulation. The National Banking Act, for example, required that banks back their banknotes
with government bonds. This limited their ability to adjust banknotes to changes in demand (note,
however, that this did provide the government with a guaranteed demand for debt; given that this was
passed during the Civil War suggests this was a defense-based regulation). As a result, changes in the
demand for banks notes would show up in the money multiplier. If the banking system did not, or could
not, respond, then this could cause fluctuations in economic activity. In the aftermath of the creation of
the Federal Reserve, if the central bank did not adjust the money supply in conjunction with changes in
money demand, this could also have had an influence on economic activity.
In other words, under the gold standard with a competitive, but highly regulated money supply it
might be the case that the deviations between money supply and money demand emphasized by Friedman
and Schwartz have additional effects on economic activity that is independent of the fluctuations in the
relative price of gold. This hypothesis would therefore seem to suggest that Friedman and Schwartz
provide an incomplete explanation of economic fluctuations. Whether or not this hypothesis is valid is
an open question. In my own work, I have examined Friedman’s ideas within the context of a structural
model and found some support (Hendrickson 2017).

6.4.3 Did the U.S. Transmit the Great Depression to the World?
You will notice from the discussion of the Depression by Friedman and Schwartz that I quoted above
that they do not give any reference to international influences. Given the international dimension to the
Great Depression, their U.S.-centric approach suggests that the United States exported the Depression
to the rest of the world. For example, Friedman and Schwartz cite the inflow of gold into the United
States and the lack of adherence to gold standard rules as evidence that the Depression was exported
to the rest of the world from the U.S. This element of the Friedman-Schwartz hypothesis seems to rely
on David Hume’s price-specie-flow mechanism. Their view seems to be that the U.S. was accumulating
gold reserves and not expanding the money supply in accordance with these supposed rules of the game.
Meanwhile, other countries must have been reducing their money supplies in response to gold outflows
and therefore importing deflation from the U.S.
Gertrud Fremlin (1985) challenges this idea. She finds that from August 1929 to August 1931, gold
reserves increased not only in the United States, but also in the rest of the world. This is consistent
with the Thompson-Glasner-Sumner theory of the Great Depression, but not Friedman and Schwartz.
Furthermore, as we discussed when we discussed McCloskey and Zecher (1976), there is no reason to think
any such “rules of the game” consistent with the price-specie-flow mechanism ever existed. Nonetheless,
even if we accept this claim that such rules did exist, the data does not support this idea. When Fremling
looks at the data she finds that currency increased in the United States as gold reserves accumulated.
While currency did not increase as much as gold reserves, it did nonetheless increase. Furthermore,
throughout the rest of the world, the currency in circulation declined despite increases in gold reserves.
In other words, gold reserve ratios were rising throughout the world and moreso in other countries than
in the United States (again, consistent with Thompson-Glasner-Sumner). Thus, it cannot be the case
that the U.S. was exporting the effects of the Depression to the rest of the world even by Friedman and
Schwartz’s own measure.
Some have been quick to dismiss Friedman and Schwartz’s explanation of the Great Depression on
the grounds that it ignores this international dimension. However, as Fremling (1985) is careful to point

16
out, the failure to consider the international aspects of the problem leaves their story incomplete, but it
does not render the entire thing false. Classical monetary theory would suggest that the inability of the
Federal Reserve to adequately adjust the supply of currency to meet demand could indeed result in (or
contribute to) deflation and a recession. The same can be said for the pre-Federal Reserve monetary
system that had various regulatory restrictions that prevented the supply of banknotes from adjusting to
fluctuations in demand. Furthermore, errors of the Federal Reserve in policy and in its role as lender of
last resort clearly did play a role in the Great Depression.

6.4.4 The Quantity Theory of Money: Implications and Importance

Even if the gold market/classical theory can better explain the Great Depression, there are potential
lessons from Friedman and Schwartz for monetary theory and policy analysis. Friedman and Schwartz
were, in part, responsible for reviving the quantity theory of money. They were perhaps early in empha-
sizing the importance of the quantity theory, but after the collapse of the Bretton Woods system and the
permanent shift to fiat money this view became very important.
In classical monetary theory, given the assumption of a competitive money supply, there was no
reason to worry about fluctuations in money demand because they would be met with corresponding
changes in the money supply. Under a fiat system, there is no such guarantee. To maintain price level
stability and to prevent policy induced fluctuations in the economy, a central bank would have to adjust
the money supply in accordance with money demand. Friedman (1989) outlined his view of the quantity
theory as primarily a theory of money demand. Individuals determine the level of real money balances
that they want to hold. The central bank determines the nominal quantity of money. If the level of real
money balances associated with this nominal money supply is consistent, then we are in equilibrium. If
the nominal supply of money is inconsistent with desired real money balances at the current price level
then the price level must adjust to get us to equilibrium. Inflation is therefore the result of an excess
supply of money and deflation is the result of an excess demand for money.
Leland Yeager (1969), and others, drew upon this view and outlined what they called the “monetary
disequilibrium view” of business cycles. According to this view, deviations between the supply and
demand for money would cause fluctuations in nominal income. Given informational frictions in the
economy, this change in nominal income would be split between prices and real output. As a result,
while monetary disequilibrium is resolved through an adjustment in the price level, such deviations
between the demand for and supply of money have real effects on the economy.
There is another important analogy here with regards to the classical theory. In the classical theory,
the price level was pinned down by the price of gold. Thus, since the money supply fluctuated with
money demand, there is no a priori reason to expect a tight relationship between the money supply
and the price level. Friedman emphasized that this point would be true if the central bank adjusted
policy correctly. In an op-ed in the Wall Street Journal, Friedman (2003) suggested that monetary policy
can be thought of like a thermostat.7 When policy is not operated correctly, nominal income will vary
wildly. This is akin to the temperature of a house with a broken thermostat when the outside temperature
changes. When the money supply is adjusted by the central bank in conjunction with changes in money
demand, nominal income will remain unchanged.
Consider the importance of this analogy. If you regressed the hourly change in the temperature of
the house on the number of minutes that the air conditioner (or furnace) was running, you would find
that marginal effect of the air conditioner is approximately zero. This is because, if the thermostat is
functioning properly, the temperature will not change. However, this is not because the air conditioner is
not having any effect on the room’s temperature. Quite the opposite, in fact. This helps us to understand
why it might be difficult to test Friedman’s theory. Suppose there is no growth in the economy and the
7
See the great blog post by Nick Rowe (2010) on this concept.

17
goal is to keep prices constant. This implies that nominal income should be constant over time. If the
central bank is optimally adjusting the money supply to changes in money demand, then we should not
expect to observe an empirical relationship between the money supply and nominal income. This is an
important lesson to keep in mind when doing reduced form empirical work.

6.5 Why Regimes Matter (Again): The Phillips Curve and the Stagflation of the 1970s
6.5.1 Some Background
In the aftermath of the Great Depression, the Keynesian Revolution took hold in the economics pro-
fession. Macroeconomics was dominated by Keynesian theories. Unfortunately, income-expenditure
Keynesianism lacked a theory of price level determination. Eventually, Keynesians settled on the idea
that inflation fluctuated in accordance with excess capacity in the economy, or lack thereof. When there
is excess capacity in the economy, the inflation rate (and perhaps the price level) will decline. Inflation
will increase as excess capacity declines.
This view was most famously encapsulated by the Phillips Curve (Phillips 1958). Broadly speaking,
what has come to be known as the Phillips Curve showed a relatively stable long-run negative relationship
between inflation and unemployment that seemed to be fairly robust across countries. Samuelson and
Solow (1960) seem to argue that this is a structural relationship. In other words, they make statements to
the effect that if unemployment is low, society will just have to tolerate inflation. Policymakers took this
a step further. Many policymakers, such as Arthur Burns, became convinced that monetary policy could
be used to increase economic activity, which would then have an effect on inflation. The optimal policy,
according to this view, was to have monetary policy aim at lowering unemployment and have wage and
price controls to prevent inflation. This is encapsulated in the following quote from Arthur Burns’s diary.
In his diary, he wrote:

What the boys that swarm around the White House fail to see is that the country now faces
an entirely new problem - namely, a sizable inflation in the midst of recession; that classical
remedies for fighting inflation or recession will simply not do; that new medicine is needed
for the new illness . . . [a] market-oriented range of policies - popularly labeled “incomes
policy” - is absolutely necessary to shorten the transitioning phase, in which we are now
caught, of moving from cost-push inflation to economic balance” (Ferrell 2010, p. 28).

Throughout Burns’s diary one comes away with the impression that Burns viewed monetary policy as a
way to influence aggregate demand, while incomes policy would take care of inflation.
The Monetarist counterrevolution revived the quantity theory of money as the determinant of the
price level. Milton Friedman’s (1968) presidential speech at the American Economic Association’s annual
meeting provides a clear explanation of the Monetarist view. Friedman’s critique of the Phillips Curve,
written in a quantity theoretic context, reversed the direction of causation.8 He implicitly assumed that
causation ran from the inflation rate to the unemployment rate. He argued, however, that this could only
be possible if inflation was unexpected. If there was a general expectation of price stability, inflation
is unexpected which means that real wages are lower than expected, which increases the demand for
labor and lowers unemployment. If inflation was fully anticipated, this would be captured in nominal
wages. Thus, the only way for inflation to have an effect on unemployment was if it was unexpected. A
policymaker who attempted to exploit the relationship would have to resort to an ever-accelerating rate
of inflation.
8
The earliest notion of a “Phillips Curve” is actually a paper by Irving Fisher in the mid-1920s. Fisher argued that causation
was running from inflation to unemployment – the opposite direction as the Keynesian view. Given that the U.S. was on the
gold standard, this is what the classical theory would predict. We will say more on this below.

18
 Robert Lucas levied his own critique a short time later that produced similar arguments to Friedman
with regards to the importance of expectations. We will now discuss Lucas’s model in more detail because
it can provide us with another idea about why the choice of framework depends on the monetary regime.

6.5.2 Lucas’s Island Model

In this section, I am going to present a simplified version of Lucas’s (1972) model.9 Consider an economy
that consists of two islands. Time is discrete and lasts forever. Everyone on these islands, however, only
lives for two periods. At any particular point in time, there are two generations of people who are living.
One group is a generation of “old people” who are in the second period of their life. The other group is
a generation of “young people” who are in the first period of their life. We will assume that each period
2/3 of the “young” generation are born on one island and 1/3 of the “young” generation are born on
the other island. Nobody knows whether or not they are on the big island or the small island. When
individuals enter old age, some of them migrate such that 1/2 of all old people are on each island. We
will assume that there are always N people in each generation.
We will assume that only young people can produce. Young people can then either consume their
production or they can sell it to old people in exchange for money. Old people will have money if they
produced when young and sold some of this production to the previous old generation. Suppose that
young people receive an endowment, y, which they can consume or combine with labor to produce. We
will assume that the young produce according to an upward-sloping supply curve. We will denote the
production of young people as `(pi ), where pi is the price on island i, where `0 (pi ) > 0. The reason that
the price can differ is because the money supply can change over time and because there are different
numbers of producers on different islands. Let’s illustrate what I mean.
The budget constraint for the young in the first period of their life is given as

pit ct + pit `t (pit ) = pit y (8)

They sell their production to old people in exchange for money. This implies that

pit `t (pit ) = mt

Thus, their second period budget constraint is

pjt+1 ct+1 = mt + pjt+1 at+1 (9)

where j is the island the individual is on when old and at+1 is a lump sum transfer received from the
central bank. It is possible that either i = j or i 6= j, depending on whether the individual moves.
Recall that the individual produces in order to obtain money to spend in old age, (pit `t (pit ) = mt ).
This implies that in the aggregate
Mt
pit N i `t (pit ) = (10)
2
where N i is the number of young people on island i and M is the aggregate money supply. In other
words, market clearing requires that prices adjust such that the aggregate quantity of money on island i
is enough to pay for the total amount of production.
Suppose that 1/3 of young agents are born on island A this period and 2/3 are born on island B. It
follows from our market clear condition that
Mt /2
pA
t = (11)
(1/3)N `t (pA
t )
9
This simplified version is based on the discussion in Champ, Freeman, and Haslag (2011).

19
 Mt /2
pB
t = (12)
(2/3)N `t (pB
t )

Proposition 1. The price on the “small island” will be higher than the price on the “big island,” or pA > pB .

Proof. Let’s assume that the reverse is true. Suppose that pA ≤ pB . Since supply is an increasing function
of price, it follows that
`(pA ) ≤ `(pB )
Multiply both sides by (1/3)N to get

(1/3)N `(pA ) ≤ (1/3)N `(pB )

We know that
(1/3)N `(pB ) < (2/3)N `(pB )
So,
(1/3)N `(pA ) < (2/3)N `(pB )
Re-arrange this to get
1 1
A
>
(1/3)N `(p ) (2/3)N `(pB )
Multiplying Mt /2 to both sides yields

Mt /2 Mt /2
A
>
(1/3)N `(p ) (2/3)N `(pB )

Which implies, by definition, that

pA > pB
However, we assumed that pA ≤ pB . This is a contradiction. Therefore it must be the case that our
assumption is wrong. This implies that pA > pB .

The intuition here is simple. When there are more producers on the island, the supply is higher and
therefore prices are lower. So why does this matter? Well, it says that prices send a signal about how
much the supplier should produce. If the individual is on the “small” island with fewer young people,
then the individual knows that the price will be higher and he/she should provide more labor. If the
individual is on the “big” island, then the individual knows that the price will be lower and he/she should
provide less labor. The basic idea is that price provides a signal about how much to produce, given
market conditions.
Nonetheless, if we look at the equilibrium condition again,

Mt
pit N i `t (pit ) =
2
what we see is that the money supply might have some impact on the price as well. However, an increase
in the money supply could also affect the price. So we might want to understand what affect monetary
policy can have on the labor market. Then we can get an idea about whether a Phillips Curve-type
relationship holds.10 To do so, we will consider two scenarios, one in which the money supply is random
and one in which the money supply grows at a constant rate over time.
10
Note that we have employment in the model, but we do not have unemployment. So we are simply looking at the Phillips
Curve-type relationship in a different way.

20
Random Money Growth. Suppose that money supply today is a multiple of the money supply yester-
day such that Mt = zt Mt−1 , where zt = 2 with probability q and zt = 1 with probability 1 − q. From
our equilibrium condition this implies that there are 4 different states of the world. Let’s call these states
A, B, C, and D:

• In state A, the money supply is constant and the individual is on the small island (z = 1 and
N i = 1/3).

• In state B, the money supply is constant and the individual is on the large island (z = 1 and
N i = 2/3).

• In state C, the money supply doubles and the individual is on the small island (z = 2 and
N i = 1/3).

• In state D, the money supply doubles and the individual is on the large island (z = 2 and
N i = 2/3).

Recall our equilibrium condition:

Mt
pit N i `t (pit ) =
2
It follows that
Mt−1 /2
pA
t = (13)
(1/3)N `t (pA
t )
Mt−1 /2
pB
t = (14)
(2/3)N `t (pB
t )
2Mt−1 /2
pC
t = (15)
(1/3)N `t (pC
t )
2Mt−1 /2
pD
t = (16)
(2/3)N `t (pD
t )

Compare the case A and case D. For example, pA and pD can be re-written as

Mt−1
pA
t =
(2/3)N `t (pA
t )

Mt−1
pD
t =
(2/3)N `t (pD
t )

Thus, pA = pD . It follows that given this monetary policy, the individual cannot tell if he/she is in state A
or state D. So the individual makes the same labor market decision when the money supply is constant
and he/she is on the small island as the individuals who is on the big island when the money supply
doubles. But recall from the proof of Proposition 1 that pA > pB and therefore pD > pB . Furthermore,
we can state the following,

Proposition 2. In the economy with random money growth, pC > pA .

Proof. We know that pA = pD . If we can show that pC > pD , then it must be true that pC > pA .
Let’s guess that pC ≤ pD . It follows that

`(pC ) ≤ `(pD )

21
Multiply both sides by (1/3)N to get

(1/3)N `(pC ) ≤ (1/3)N `(pD ) < (2/3)N `(pD )

Re-arranging yields
1 1
D
<
(2/3)N `(p ) (1/3)N `(pC )
Multiply both sides by Mt−1 to get

Mt−1 Mt−1
D
<
(2/3)N `(p ) (1/3)N `(pC )

By definition, this means that

pD < pC
This contradicts our assumption that pC ≤ pD . Therefore our assumption must be wrong. It follows that

pC > pD

So Propositions 1 and 2 imply that pB < pA = pD < pC , which also implies that `(pB ) < `(pA ) =
`(pD ) < `(pC ). But note that pC is associated with z = 2 and pB is associated with z = 1. It follows
that if you were to plot z and `, you would find a positive relationship. In other words, you would get a
positive relationship between inflation and employment. This is the Phillips Curve.
But what happens if we assume that the money supply grows at a constant rate over time?

Constant Money Growth. Now suppose that instead of assuming that money grows randomly over
time, we assume that Mt = zMt−1 , where z is some constant. So what are the implications from this
assumption? Well, recall the budget constraints when young and when old and the equation that reflects
the exchange of money for production:

pit ct + pit `t (pit ) = pit y

pit `t (pit ) = mt
pjt+1 ct+1 = mt + pjt+1 at+1
Combining these we get
pit ct + pjt+1 ct+1 = pit y + pjt+1 at+1

Dividing through by pjt+1 and solving for ct+1 yields

pit pit
ct+1 = (y − ct ) + at+1 = `t + at+1
pjt+1 pjt+1

Thus, the slope of the budget line is the inverse of the (gross) inflation rate. So what determines inflation
in our model. Well, we can use the equilibrium condition, equation (), again to write:
Mt+1
pjt+1 N i `t (pit ) Mt+1 N i `t (pit )
 
2N i `t (pjt+1 )
= = = z (17)
pit Mt
2N i `t (pit ) N i `t (pjt+1 ) Mt N i `t (pjt+1 )

22
Note two things. Since the money supply is process is known, individuals incorporate it into their
decision-making process. It follows that there is no uncertainty about which island the individual is on.
So the higher the rate of money growth, all else equal, the higher the rate of inflation, and the flatter then
budget line becomes. It follows that the individual will produce less in period 1. Why? Because inflation is
a tax on currency. Thus, the individual will want to economize on currency, which implies that production
declines. This implies that higher rates of inflation are associated with lower levels of employment. This
is the opposite implication of the Phillips Curve. (Note that our previous result was a time series results
and this is a cross-sectional result. Nonetheless, this example implies that cross-sectional evidence should
show that higher rates of inflation are associated with higher unemployment.)

Implications. What the model implies is that the monetary regime is important. When the money
supply process is known with certainty, it is only the size of the island that matters and money does
not have any effect on production decisions. However, when the money supply process is random, this
creates confusion, which causes changes in the production decisions of the individuals of the model.
This is important because the model implies that Phillips Curve-type relationships only emerge when the
money supply is randomly changing. This is consistent with Friedman’s idea in his 1968 speech.

6.5.3 What We Can Learn About the Phillip’s Curve

Lucas (1976) generalized this basic idea in what is now known as the Lucas Critique. Lucas argued that
reduced form econometric models were not capable of policy analysis. The reason that they were not
useful is that the parameter estimates were based on historical data. As a result, the coefficient estimates
represented the relationship between the variables under a particular policy regime. If the policy regime
changes, then the coefficient estimates might change as well. He argued that in order to understand the
effects of changes in the policy regime, we would need to understand the deep, structural parameters
that underlie this estimates. In other words, when policy regimes change, people in the economy tend
to change their behavior. The estimated relationships between variables would change and previous
estimates would not be useful for policy analysis.
We can use the Lucas model discussed above to understand the importance of regimes. However,
rather than focus on the model itself, let’s return to our lesson about different monetary regimes. As
should now be quite clear, our general view is that under the gold standard, the appropriate model for
thinking about monetary effects on economic activity is the classical theory. When money is inconvertible
paper money, the appropriate model is the quantity theory. Let’s apply this insight to the Phillips Curve.
According to the classical theory, fluctuations in the price level are driven by changes in the supply
of and demand for gold. Many of these changes are unexpected. If nominal wages are slow to adjust,
as the classical economists seem to believe (see Thornton 1802, for example), then changes in the price
level would cause an opposite change in real wages. In other words, when prices would rise unexpectedly,
real wages would decline. The profitability of firms would increase and therefore so would production
and employment. When prices would decline unexpectedly, real wages would rise and the profitability of
firms would decline and therefore so would production and employment. Thus, under the gold standard,
we see a negative relationship between inflation and unemployment that is consistent with the Phillips
Curve.
With an inconvertible paper money, like during the 1970s, the appropriate model is the quantity
theory. Anticipated higher rates of money growth cause higher rates of inflation. Since inflation is a
tax on currency, individuals economize on currency and consumption, output, and employment decline.
This implies that not only does the Phillips Curve no longer describe the data, but that the slope of an
estimated Phillips curve would have the opposite sign! As Lucas shows, only if changes in the money
supply are random can you generate data that looks like the Phillips Curve.

23
 The lesson here is that the empirical relationship between inflation and unemployment encapsulated
by the Phillips Curve is regime-dependent. Under the gold standard, when changes in the price level are
random and wages are sticky, you can estimate a relationship between inflation and unemployment that
makes it appear as though there is a trade-off between the two. However, under a fiat money system, the
only way to replicate this empirical pattern is to have the money supply change randomly over time. If the
central bank makes any attempt to exploit this possible trade-off, then people will expect higher inflation
whenever unemployment is high and the central bank will be unsuccessful at reducing unemployment.

7 How Do We Measure the Money Supply? (And Does It Matter?)

If we want to examine the role of money in the macroeconomy, it might be useful to consider issues of
measurement. For example, think about how we calculate real GDP. What we want to measure is total
production. However, this is difficult to do because production lacks a common unit of measurement.
How do you add 2 bushels of wheat, 3 gallons of milk, and 1 airplane to get an aggregate quantity of
production? You should know that the answer is not to say that production is 6 units. With GDP, we
use money prices as the common unit of measurement. We can then calculate the total value of the
production of each good and add these up to get the total value production. This nominal measure of
production is not useful if we are trying to distinguish between trends in production and trends in prices.
Nominal GDP could be rising because prices are rising, production is rising, or both. If we are interest in
production, we need to isolate prices from quantities. Thus, we use some technique like chain-weighting
in order to hold prices constant and get an estimate that only captures changes in quantities over time.
Friedman and Schwartz (1970) recognized that this same issue might apply to money.11 For example,
when we say “money”, we could be referring to a multitude of assets. Currency, checking accounts,
savings accounts, time deposits, etc. are all referred to as “money.” What Friedman and Schwartz
pointed out was that while each of these things is money, they differed in their degree of “moneyness.”
What they meant is that the flow of services from holding money differs depending on the type of money
that one is holding. So why does this matter? It matters because the money supply has typically been
measured by simply adding up all of the different types of money using simple summation. According to
Friedman and Schwartz, each type of money should not simply be given equal weights, but rather each
asset should be weighted by its degree of “moneyness.”
A different way of putting this is that each asset should have value-based weights, like any other
index. William Barnett (1980) derived monetary aggregates using standard consumer choice theory.12
These alternative monetary aggregates are referred to as Divisia monetary aggregates because they draw
on the work on index numbers by the French Mathematician Francois Divisia. The Divisia aggregates
can be written as13
n  
X 1
∆ ln Mt = (ωi,t − ωi,t−1 ) × ∆ ln(xi,t ) (18)
2
i=1
11
Irving Fisher discussed this even earlier.
12
Others have proposed alternatives aggregates as well. Rotemberg, Driscoll, and Poterba (1995), for example, proposed
something called a “currency equivalent,” which is
n
X
CEt = [(Rt − rit )/Rt ]xi,t
i=1

where R is a benchmark rate and ri is the rate of return on the particular asset. This is a weighted average of monetary
components as well, but there is one critical difference. The Divisia aggregate is a flow of services index. The CE is a weighted
average of money stocks.
13
We will not derive this here, but you can refer to Barnett (1980) or the textbook treatment by Serletis (2007).

24
where xi,t is the flow of services from a monetary asset that is assumed to be proportional to the stock
of the asset, ωi,t is the weight of asset i in the index and is defined as
pi,t xi,t
ωi,t = Pn (19)
i=1 pi,t xi,t

where pi,t is the user cost of money derived by Barnett (1978):

 
Rt − ri,t
pi,t = Pt (20)
1 + Rt

where R is a benchmark rate of return on non-monetary assets, rit is the rate of return on the asset, and
Pt is the price level. These aggregates are explicitly derived from economic theory and are consistent
with index number theory. By contrast, traditional monetary aggregates are simple sum indices. For
these indexes to be consistent with economic theory and index number theory, it would have to be true
that all monetary assets are perfect substitutes with one another. This idea, however, does not pass
rigorous empirical testing nor does it pass a basic “smell test.” These characteristics have implications
for anyone who wants to examine the role of money in a macroeconomic context, then one must be sure
that money is being measured properly. If someone uses simple sum aggregates and gets results that
seem at odds with monetary theory, then this suggests that either the model is wrong or the improper
aggregation technique is driving the results.
This issue is important because the importance of money for monetary policy and macroeconomics
generally has recently been called into question. Some economists have argued that money is no longer
useful in predicting inflation or that changes in real money balances could explain fluctuations in real
economic activity. This latter effect is important because Monetarists often argued were important to
understanding the transmission process of monetary policy because they served as a proxy for an index
of portfolio effects. Mike Belongia (1996) and I (Hendrickson 2014) have shown that these results do not
hold up to scrutiny. These puzzling empirical results appear only when one uses simple sum monetary
aggregates. When using Divisia monetary aggregates, the results are consistent with standard quantity
theoretic arguments and the Monetarist transmission mechanism.

8 Money and Conventional Models

By this point, you should be fairly convinced that money plays an important role in the macroeconomy.
But suppose we want to do some sort of quantitative analysis or generate some type of policy implication.
To do that we need to have some sort of economic model. If we think about standard Arrow-Debreu-type
general equilibrium models, there is not really a role for money. These models do a good job explaining
how prices work to coordinate resources, but there is no process of exchange. Prices simply adjust to
clear all markets. It is as though there is some sort of auctioneer operating in the background, exploring
various price vectors until excess demand is zero in every market.
This discussion suggests that if you are interested in writing down a monetary model, then you need
to take into account the exchange process in that model. In addition, it would be potentially advantageous
to introduce money into a model that we are all familiar with. Unfortunately, a lot of the literature has
focused on the latter and not the former. In fact, the way in which many people have attempted to add
money into the model is by introducing money into the utility function or by adding cash-in-advance
constraints. These seem ill-equipped to address the important properties that I have highlighted above.
Nowhere in my discussion of why money was important or how money evolved did I ever mention
anything that could be construed as saying that money exists because people get satisfaction just from

25
holding it.14 Rather, the benefit from using money is that it facilitated trade that never would have taken
place otherwise. Cash-in-advance constraints similarly make little sense in this context since they imply
that money imposes an additional constraint on decision-making and therefore money actually makes
people worse off.
Advocates of these approaches might suggest that these modeling techniques are just simplifications
to get money into the model. This would be fine if these simplifications captured some degree of
the importance of money in our discussion above – the cash-in-advance constraint being the likeliest
winner since it does capture that empirical fact that goods trade for money and money trade for goods.
Nonetheless, Woodford (2003) has demonstrated that these sorts of monetary models can be solved
without money and the implications of the model do not change. In quite the turn of events, many
economists have convinced themselves that if money doesn’t matter in their model, then it must not
matter in the real world! Let me summarize this bit for emphasis. First, economists have taken models
in which money is not important and added money in a way that does not capture the importance of
money. Second, they have solved these models and found that money turns out to be unimportant in
the model. Third, they have decided that money is not important in the real world because it is not
important in the model. This is circular reasoning.
We will ultimately get to these “moneyless” models. For now, however, I want to introduce two
ways of trying to incorporate money into the model.15 One involves transactions costs and is a modest
departure from conventional models. In this model, in a monetary equilibrium, money is a substitute to
storing goods over time when storing goods is costly. The other involves search and matching and is an
extension of the model we used in section 3 to include divisible money and divisible goods.

9 Money and Transaction Costs

Early attempts to integrate money and value theory relied on transaction costs to get money into the
model. In this section, I present a continuous time version of Niehans’s (1979) transaction cost model of
money and discuss the implications.
We will assume that individuals have expected lifetime utility,
Z ∞
E0 e−ρt u[c(t)]dt
0

where ρ is the rate of time preference, c is consumption, and u0 , −u00 > 0.

The individual receives income in the form of an endowment of goods, x. These goods can be
consumed, stored over time, or sold for money. Let sg denote the stock of goods owned by the individual.
It follows that the stock of goods evolves according to:

dsg = [x − c + z − y − γsg − ω(z + y)]dt (21)

where z is the purchase of new goods, y is the sale of goods to others, γ > 0 is the cost of holding goods
over time, and ω is the transaction cost associated with buying and selling goods. One can think of the
storage costs and the transaction costs associated with buying and selling as simply destroying some of
the stock of goods.
14
Money in the utility function is what I call the “Scrooge McDuck Theory of Money”, named after the cartoon character
created by Carl Banks in the 1940s. During my childhood, Scrooge McDuck was one of the main characters on a show called
DuckTales and he would dive into and subsequently swim in a pit of gold coins, which seems to violate the laws of physics.
Nonetheless, this seems to the only person/duck for which money in the utility function is a good starting point.
15
Our overlapping generations model of the previous section was one such model, but it may or may not be useful for policy
analysis. This depends on how you want to interpret it. See Tobin (1980) for a critical comment. See Andolfatto (2015) for an
example of how to use the model to discuss policy.

26
 Money evolves according to
dsm = [y − z + rsm + τ ]dt (22)
where sg is the stock of real money balances, r is the rate of return paid on money, and τ is a lump-sum
transfer/tax from the monetary authority.
Let V (sg , sm ) denote the value function. The Bellman equation for the individual can be written as

ρV (sg , sm ) = max u(c) + Vsg [x − c + z − y − γsg − ω(z + y)] + Vsm [y − z + rsm + τ ] (23)
c,y,z

The first-order conditions are given as

u0 (c) = Vsg (24)
(1 − ω)Vsg = Vsm (25)
(1 − ω)Vsg = Vsm (26)
Note from the first first-order condition that the marginal utility of consumption is equal to the marginal
value of holding an additional unit of the good over time. This should be fairly intuitive since the
individual receives an endowment of goods, if u0 (c) − Vsg > 0, then the individual gets a greater benefit
from consuming an additional good than from saving and storing that good.
Combining the first-order conditions yields

(1 − ω)u0 (c) = Vsm (27)

’ This suggests that the marginal utility of consumption, adjusted by the marginal cost of trading goods,
is equal to the marginal value of holding an additional unit of money.
From the Bellman equation,
ρVsm = rVsm + Vsm sm s˙m
ρu0 (c) = ru0 (c) + u00 (c)ċ
u00 (c)
− ċ = r − ρ
u0 (c)
This is a familiar condition from our consumption model. But there is a new wrinkle here, which we
need to explain. First, however, consider that if we differentiate the value function with respect to sg , we
get
ρVsg = −γVsg + Vsg sg s˙g
Or,
u00 (c)
− ċ = −(γ + ρ)
u0 (c)
If goods and money are held in equilibrium it must be true that

r = −γ

So goods will be held over time if the rate of return on money is negative and equal to minus the marginal
cost of holding goods. Recall from the Fisher equation that i = r + π, where i is the nominal interest
rate, r is the real interest rate, and π is the inflation rate. Since money does not pay interest, this implies
that r = −π. Re-writing the condition above, it follows that individuals will hold goods over time if

π=γ

27
This is fairly intuitive since it suggests that individuals will hold both money and goods over time if the
marginal cost of holding money is equal to the marginal cost of storing goods.
However, for a stationary equilibrium to exist in our economy, if must be true that ċ = 0. From our
equilibrium conditions, this implies that both of the following conditions must hold:
π = −ρ
γ = −ρ
This implies that in any stationary equilibrium, the rate of inflation will be equal to minus the rate of
time preference. Furthermore, the storage cost of holding goods must be equal to the rate of inflation.
However, this would require the storage cost to be negative, which violates our assumption that storage
costs are positive (it is hard to imagine why storage costs would ever be negative). Since the storage cost
cannot be negative, this implies that money, but not goods will be held in a stationary equilibrium.
Finally, note that π = −ρ implies that r = ρ. So the rate of return on money is equal to the rate
of time preference. This is a result that is known as the Friedman Rule. We will discuss this at a much
greater length below. Nonetheless, a significant limitation of this model since it implies that a stationary
monetary equilibrium can only exist if the Friedman Rule holds. We might be able to resolve this by
amending the model (as Niehans himself did) by adding bonds and other assets as well as other potential
frictions.

10 Monetary Search with Divisible Goods and Divisible Money

Another method of getting money into models is through a search and matching framework. We in-
troduced this sort of idea above when we discussed the network effect associated with holding money.
However, in that model we had indivisible goods and indivisible money. If we want these sort of models
to be useful for analyzing issues of monetary policy and the interaction between money and financial
markets, then we need to have divisible goods and divisible money. In this section, I present the Lago-
Wright (2005) monetary search model in both discrete and continuous time.16 I work through the discrete
time version of the model to discuss the conditions under which a monetary equilibrium would exist. I
then use the continuous time version of the model to discuss the policy implications.

10.1 A Model in Discrete Time

Time is discrete and infinite. We will assume that there is a continuum of individuals with unit mass
(i.e. the population is normalized to 1). All of the individuals in our economy are assumed to be
identical. These individuals participate in two markets each period. Thus, we will divide each period
into two distinct subperiods. In the first subperiod, individuals are matched pairwise to trade in what
will hereafter be referred to as the decentralized market. In the second subperiod, individuals trade in
a Walrasian-style market that will hereafter be referred to as the centralized market. Individuals can
produce a distinct good in each market, the decentralized market good and the centralized market good.
These goods are not storable over time. When individuals are matched pairwise to trade, these meetings
are anonymous. The anonymity of meetings implies that credit is infeasible and therefore a medium of
exchange is necessary. We will assume that there is an intrinsically useless, inconvertible paper money in
existence that can possibly serve as a medium of exchange.
The expected lifetime utility of individuals in our model is
∞
X
U = E0 β t [u(qtb ) − c(qts ) + xt − ht ] (28)
t=0
16
For a textbook treatment of the model, see Nosal and Rocheteau (2011).

28
where β ∈ (0, 1) is the discount factor (i.e. 1/(1 + ρ) where ρ is the rate of time preference), q b is the
quantity consumed in the decentralized market if the individual is a buyer, q s is the quantity produced
in the decentralized market if the individual is a seller, x is the quantity of the centralized market good
that the individual consumes, h is the number of hours worked in the centralized market to produce
the centralized market good, u is a function with the properties u0 , −u00 > 0, and c is a function that
measures the disutility of production such that c0 > 0, c00 ≥ 0. We will assume that the individual
produces an amount y in the centralized market according to the linear production function yt = ht .
Finally, we will assume that there is a central bank that adjusts the money supply through lump sum
payments.

10.1.1 The Centralized Market

An individual in the centralized market produces yt = ht units of the centralized market good. Together
with existing money balances and a real lump sum transfer from the central bank, τ , the individual uses
this production to consume the centralized market good. Whatever is not consumed is held as money
and carried over into the next period. The evolution of money balances in the centralized market can be
written as
φt mt+1 = φt mt + yt + τt − xt (29)
where φt is the price of money in terms of the centralized market good and mt is money.
Let W (mt ) denote the value associated with carrying money balances, mt , into the centralized
market. It follows that
W (mt ) = max xt − yt + βEt V (mt+1 ) (30)
xt ,yt

subject to m0 = m̄ and equation (29), where V (mt ) is the value function in the decentralized market
and E is the expectations operator. Substituting the budget constraint into the value function, we can
now write
W (mt ) = max φt mt + τt − φt mt+1 + βEt V (mt+1 ) (31)
mt+1

The first-order condition is given as

φt = βEt V 0 (mt+1 ) (32)
This condition implies that the price of money is equal to the expected discounted marginal value of
money in the next period’s decentralized market.

10.1.2 The Decentralized Market

Individuals begin each period in the decentralized market. In this market, the individuals are matched
pairwise to trade. We assume that individuals are buyers who are matched with a seller with probability
α, sellers who are matched with a buyer with probability α, and unmatched with probability 1 − 2α.
When individuals are matched, they negotiate the terms of trade. An individual who is holding m units
of money (“dollars”), will offer d ≤ m dollars in exchange for the good produced in the decentralized
market. We need to be more specific about the terms of trade. For simplicity, we will assume that
buyers make take-it-or-leave-it offers to sellers. For this offer to be feasible, it must satisfy the following
constraint:
φt d ≥ c(qts ) (33)
where
d≤m (34)

29
This condition essentially says that the real value of money balances offered to the seller must be greater
than or equal to the disutility the seller gets from producing. Furthermore, the buyer cannot offer more
dollars than they actually have. It follows that the buyer’s problem is given as

max[u(qtb ) − φt d] (35)
qt ,d

subject to
φt d ≥ c(qts ) (36)
and a market-clearing condition:
qb = qs (37)
We can write the Lagrangian as

max[u(qt ) − φt d] + µ[φt d − c(q)] + λφt [mt − d] (38)

qt ,d

The Kuhn-Tucker conditions are given as

u0 (qt ) = µc0 (qt ) (39)

1+λ=µ (40)
Combining this conditions yields
u0 (qt ) = (1 + λ)c0 (qt ) (41)
Note that if λ = 0, then d < m. This means that u0 (q) = c0 (q). If λ > 0, then d = m and u0 (q) > c0 (q).
The feasibility constraint will always be binding when trade takes place. For now, we will assume that
d = m and we will return to it later to see if this is a safe assumption. It follows from this assumption
that φt m = c(qt ).
Let V (mt ) denote the value function of an individual who enters the decentralized market with
money balances, mt , and W (mt ) denote the value function of an individual who subsequently enters the
centralized market with money balances, mt . Given our assumptions, we can write the value function for
the individual entering the decentralized market as

V (mt ) = α[u(qt ) + W (mt − d)] + α[−c(qt ) + W (mt + d)] + (1 − 2α)W (mt ) (42)

Given the linearity of W (mt ), it follows that this can be rewritten as

V (mt ) = α[u(qt ) − c(qt )] + W (mt ) (43)

Differentiating this with respect to mt yields

∂qt
V 0 (mt ) = α[u0 (qt ) − c0 (qt )] + W 0 (mt ) (44)
∂mt
From the implicit function theorem,
∂qt φt
=
∂mt c(qt )
Substituting this in the differentiated function, intreating the expression forward, and substituting it into
our first-order condition from the centralized market gives us:
 0 
φt u (qt+1 )
Et − 1 = αEt 0 −1 (45)
βφt+1 c (qt+1 )

30
We can now circle back to our discussion of the bargaining problem. Recall that from the bargaining
problem we had u0 (q) = (1 + λ)c0 (q). This implies that we can re-write our equilibrium condition as
φt
Et − 1 = αλ (46)
βφt+1
In order for λ to be positive, the left-hand side must be positive. This tells us that d = m if φt > βφt+1
and d < m if φt = βφt+1 . Note, however, that if φt < βφt+1 this implies that the right-hand side of our
equilibrium condition would be negative. This would imply that the marginal utility from consumption
is less than the marginal disutility of production. If this is the case, however, then no trade would take
place. Thus, a monetary equilibrium requires that φt ≥ βφt+1 . So what does this mean? Note that
there are not frictions in our model. So money is neutral, which implies that real money balances are
constant over time, i.e., φt+1 Mt+1 = φt Mt . Let’s suppose that the central bank maintains a constant
money growth rule, Mt = µMt−1 . It follows that for a monetary equilibrium to exist, it must be true
that µ ≥ β. We will discuss what this means for optimal monetary policy in the next section.

10.2 A Model in Continuous Time

Now let’s consider the same model in continuous time. Before we get to the model, we should make
note of an important difference we have to make. Since we are in continuous time, each “period” of
time is infinitesimal. Thus, it doesn’t make much sense to think about each period being divided into
two subperiods. So what we will do is assume that individuals basically live in the centralized market,
but randomly get matched pairwise to trade. For continuity’s sake, we will continue to refer to pairwise
trade as the decentralized market. Nonetheless, you could simply assume that individuals occasionally
just bump into somebody to trade with in a pairwise fashion while they are in the centralized market.
Regardless, this formulation gives us something tractable we can work with that resembles our discrete
time model. Specifically, this allows us to think about the centralized market value function as referring to
continuous trading in the centralized market up to the point when the individual enters to decentralized
market. Once trade is concluded in these pairwise meetings the individual immediately returns to the
centralized market.
A continuous time representation of expected lifetime utility is
Z ∞
U =E e−ρt {u[q b (t)] − c[q s (t)] + x(t) − y(t)}dt (47)
0

where the variables are defined the same as in our discrete time case.
The continuous time evolution of real money balances is

dz(t) = [y(t) − x(t) + τ (t) − πz(t)]dt (48)

where z is real money balances, π is the rate of inflation, and τ is a lump sum transfer from the central
bank.
The individual in our model will be assumed to be in the centralized market where he or she
remains until randomly assigned to the decentralized market. Let F (t) denote the cumulative distribution
function that measures the probability that the individual has entered the decentralized market by time t
and f (t) denote the probability density function that measures the probability that the individual enters
the market during the interval [t, t + dt). It follows that the value function for an individual in the
centralized market can be written as
Z
W (z) = max e−ρt {[1 − F (t)][x(t) − y(t)] + f (t)V (z)}dt (49)
x(t),y(t)

31
subject to equation (48) and z(0) = z0 .
Let’s assume that the time at which the individual enters the decentralized market follows an expo-
nential distribution with an arrival rate, α. It follows that

F (t) = 1 − e−αt

f (t) = αe−αt
This implies that the value function now satisfies
Z
W (z) = max e−(ρ+α)t {[x(t) − y(t)] + αV (z)}dt (50)
x(t),y(t)

subject to equation (48) and z(0) = z0 .

It follows that we can write the HJB equation:

(α + ρ)W (z) = max x − y + αV (z) + W 0 (z)(y − x + T − πz) (51)

x,y

The first-order conditions for maximization are redundant. The condition is

W 0 (z) = 1

Differentiating W (z) yields

(α + ρ)W 0 (z) = αV 0 (z) + W 00 (z)dz − πW 0 (z) (52)

From our first-order condition, we know that W 0 (z) = 1 and W 00 (z) = 0. Therefore, this simplifies to

α + ρ + π = αV 0 (z) (53)

When individuals enter the decentralized market, we assume that they are a buyer who is matched
with a seller with probability γ, a seller matched with a buyer with probability γ, and not matched with
probability 1 − 2γ. When two individuals are matched pairwise to trade, they must trade goods for
money and money for goods because trade is anonymous. If they are not willing to accept money, then
no trade takes places because it is the only asset in our model. We will assume that buyers will offer
a quantity of real money balances, d ≤ z to the seller. Finally, once trade is completed the individuals
return to the centralized market. These assumptions imply that we can write the value function in the
decentralized market as

V (z) = γ[u(q b ) + W (z − d)] + γ[−c(q s ) + W (z + d)] + (1 − 2γ)W (z)

Or, re-arranging,

V (z) = γ[u(q b ) − c(q s )] + γ[W (z − d) + W (z + d)] + (1 − 2γ)W (z)

Note that W (z − d) + W (z + d) = 2W (z) because of the linearity of W . So this reduces to

V (z) = γ[u(q b ) − c(q s )] + W (z) (54)

Differentiating with respect to z we get

b s
 
0 0 ∂q 0 ∂q
V (z) = γ u −c + W 0 (z) (55)
∂z ∂z

32
So, we need to understand the terms of trade to complete our model. We will assume that buyers make
take-it-or-leave-it offers to sellers such that

d ≥ c(q s )

d≤m
We will assume that the second constraint is binding (we can check the conditions for this later). Fur-
thermore, will we assume that buyers never off to pay any amount more than they have to in order to
get sellers to trade. This implies that the first constraint is binding as well. Finally, we know that in any
trade, q s = q b . Thus, we can re-write the terms of trade as

z = c(q) (56)

Using the implicit function theorem,

∂q 1
= 0 (57)
∂z c (q)
Thus, equation (55) can be re-written as

u0 (q)
 
0
V (z) = γ 0 −1 +1 (58)
c (q)
Substituting this back into equation (53) yields

u0 (q)
 
ρ+π = αγ × − 1 (59)
| {z } |{z} c0 (q)
Opportunity Cost of Holding Money Probability of Trade in DM | {z }
Benefit from Trade

This tells us that the opportunity cost of holding money is equal to the expected net benefit of holding
money. In other words, money allows the individual to make transactions that he or she wouldn’t be able
to make otherwise. This is an exchange premium, or as it is sometimes called, a liquidity premium. Let
e(q) be the exchange premium. It follows that

e(q) = ρ + π = i

where i is the nominal interest rate. [Note that i = ρ + π via the Fisher equation.] So the marginal benefit
is holding an additional dollar is the exchange premium and the marginal cost is the foregone interest
represented by the market’s nominal interest rate on a substitutable asset. Recall that the nominal interest
rate paid on currency is zero (currency does not earn interest). So you might wonder why people would
ever hold any of their wealth in currency. The answer is that money provides a non-pecuniary rate of
return, which is captured by the exchange, or liquidity, premium that currency provides. This insight
will play an important role when we introduce other assets into our model. For example, something like
bank deposits might also be used in transactions. Unlike currency, bank deposits often pay interest. In
addition, we know that bank deposits are used as a medium of exchange. But why do currency and bank
deposits circulate alongside one another? Our model suggests that competing media of exchange with
different rates of return must have different exchange, or liquidity, premia. Thus, we must ask ourselves
why. Before we do, however, we will take a detour through the asset pricing literature.
There is one final point to make regarding the model. In his paper, “The Optimum Quantity of
Money,” Milton Friedman made the argument that since individuals face the cost of foregone interest
when they hold currency, individuals will tend to economize on holding currency. From a societal
perspective, positive nominal interest rates imply an inefficiency: people aren’t holding enough money.

33
Friedman suggested two ways that this could be resolved. First, central banks could pay interest on
currency. This is likely to be fairly costly, however. Thus, his second suggestion was that central banks
could lower the nominal interest rate to zero by producing deflation. Recall from our consumption
lectures that, in equilibrium, the real interest rate is equal to the rate of time preference. From the Fisher
equation, the real rate of return on currency is minus the rate of inflation (since the nominal interest rate
on currency is zero). It follows if the central bank created a policy of deflation such that the inflation rate
was equal to minus the rate of time preference, then assets would have a nominal interest rate of zero just
like money. If the nominal interest rate is equal to zero then the opportunity cost of holding currency is
zero and individuals will hold the optimal amount of currency.
Now refer back to our equilibrium condition. According to our equilibrium condition, e(q) = i. So if
i = 0, then there is no exchange premium on currency. But is this optimal? Well, let’s think about this.
When individuals meet pairwise to trade, the total possible surplus is u(q) − c(q). Suppose that we want
to maximize the possible surplus. Choosing q to maximize the surplus gives us:

u0 (q) = c0 (q)

Now let’s plug this into our equilibrium condition:

 0 
u (q)
ρ + π = αγ 0 − 1 = αγ[1 − 1] = 0
c (q)
The only way in which ρ + π can equal zero is if π = −ρ. This is the Friedman Rule!
This characteristic is not unique to our model. In fact, the conclusion that optimal monetary policy is
characterized by the Friedman Rule is a robust result across many model specifications. The robustness
of this conclusion is important because understanding the reasons why the Friedman Rule might not hold
could be important for monetary policy debates.
Nonetheless, it is important to note that the Friedman Rule corrects for one friction that exists in the
economy, the lack of interest-bearing currency. In reality, there are other possible frictions that might
lead to inefficient outcomes that are not included in our simple model. If so, then policy might have to
balance the costs and benefits of the Friedman Rule in light of these other frictions. I have examined the
ideas of monetary disequilibrium theory in the context of a basic search model when the probability of
trade is stochastic. I find that there are two types of rules that are optimal. A Friedman Rule is optimal.
However, if the Friedman rule is not feasible, an alternative objective is to minimize deviations around
the steady state. Under this objective, a nominal GDP target is optimal (Hendrickson, 2015).

11 The Coexistence of Currency and Other Assets

In the previous sections we discussed asset pricing where the assets all paid a positive expected rate of
return. Money, since it does not pay any nominal rate of return, pays a real rate of return equal to minus
the rate of inflation. In a world where other assets do pay a nominal rate of return, why would anyone
hold money? To examine this question, we will begin by assuming that there are transactions costs
associated with using any asset as a medium of exchange in the context of our continuous time monetary
search model from the previous section. We will then examine how transaction costs help explain the
rate of return dominance puzzle and then discuss what might explain why such transaction costs exist.
The basic setup of our model is the same as in the previous lectures on money. We will assume that
individuals are in the centralized market and they probabilistically bump into others for pairwise trade.
We will also assume that there are two assets, currency and bonds. Both can potentially be used as media
of exchange, but there are transactions costs associated with using them. Specifically, we will assume that
when a buyer provides a seller with currency or bonds, the seller cannot costlessly determine whether the

34
asset is authentic or counterfeit. Each individual does possess a technology that can determine whether
the asset is authentic. However, using this technology will destroy some fraction of the asset. We will
show how differences in transactions costs can explain the rate of return dominance puzzle (i.e. why
individuals hold currency alongside interest-bearing assets). We will also see why this has important
implications for monetary aggregation.17

11.1 The Centralized Market

The evolution of real money balances, z, can be written as

dz = [y + rb − x − i + τ − πz]dt (60)

where r is the real rate of return on bonds, b is real bond balances, i is the net purchases of bonds, τ is
a lump sum transfer from a central bank, and π is the rate of inflation.
The evolution of bond balances can be written as

db = idt (61)

It follows that the HJB equation can be written as

(ρ + α)W (z, b) = max x − y + αV (z, b) + Wz (z, b)[y + rb − x − i + τ − πz] + Wb (z, b)i (62)
x,y,i

where V (z, b) is the value of carrying real money balances and real bond balances into pairwise meetings.
The first-order conditions can be written as

Wz (z, b) = 1 (63)

Wb (z, b) = Wz (z, b) (64)

Differentiate the value function with respect to z and b to get

dz
(ρ + α)Wz (z, b) = αVz (z, b) − πWz (z, b) + Wzz + Wbz (z, b)i (65)
dt
dz
(ρ + α)Wb (z, b) = αVb (z, b) + rWb (z, b) + Wbb i + Wzb (z, b) (66)
dt
From the first-order conditions, Wzz = Wbb = Wzb = 0 and Wb = Wz = 1. These conditions can
therefore be re-written as
ρ + α + π = αVz (z, b) (67)
ρ + α − r = αVb (z, b) (68)
Each of these conditions now show the marginal cost of holding each asset must be equal to the expected
marginal value of that asset in pairwise meetings. Given that there are different marginal costs, it must
be the case that there are different marginal values associated with each asset. Otherwise, individuals
would only hold the asset with the lower marginal cost.
17
What follows borrows liberally from my paper “Monetary Search, Proportional Transaction Costs, and the Currency
Equivalent Index.”

35
11.2 Pairwise Meetings
When individuals bump into each other, we assume that individuals will want to purchase the decentral-
ized market good and be matched with a seller with probability γ, to produce the decentralized market
good and be matched with a buyer with probability γ, and not have any mutual gains from trade with
probability 1 − 2γ. This implies that we can write the value function as
V (z, b) = γ[u(q b ) + W (z − dz , b − db )] + γ[−c(q s ) + W (z + dz , b + db )] + (1 − 2γ)W (z, b) (69)
When they are matched, they negotiate the terms of trade. Of course, in any arrangement, it must
be the case that q b = q s = q. Again, for simplicity, assume that buyers make take-it-or-leave-it offers
to sellers. However, we will make the following claim. We will argue that there are transaction costs
associated with exchange. Specifically, we will assume that the fear of counterfeiting will encourage
sellers to inspect the media of exchange. In the process, (1 − χz ) of real balances and (1 − χb ) of bonds
will be destroyed. For the offer to be incentive compatible,
χz dz + χb db ≥ c(q) (70)
where dz ≤ z and db ≤ b.
The buyer’s problem is to choose q, dz , and db to maximize
L = u(q) − χz dz − χb db + λ1 [χz dz + χb db − c(q)] + λ2 [z − dz ] + λ3 [b − db ] (71)
The first-order conditions are given as
u0 (q) = λ1 c0 (q) (72)
(λ1 − 1)χz = λ2 (73)
(λ1 − 1)χb = λ3 (74)
Let’s go through these conditions. Subtract c0 (q) from both sides of the first condition and then divide
by c0 (q). This yields:
u0 (q) − c0 (q)
= λ1 − 1
c0 (q)
Note that for trade to take place, there has to be gains from trade. Furthermore, since the total surplus
from trade is u(q) − c(q), it follows that the surplus is maximized when u0 (q ∗ ) = c0 (q ∗ ). This implies
that when trade takes place, it should be the case that u0 (q) ≥ c0 (q). This, in turn, implies that λ1 ≥ 1.
Two results follow. First, the take-it-or-leave-it offer is always binding because λ1 > 0. Second, if λ1 = 1,
then we end up with the optimal allocation. If λ1 > 1, then λ2 > 0 and λ3 > 0, and therefore z = dz
and b = db , respectively. Intuitively, when λ1 = 1, there is an abundance of assets. As a result, buyers
will not have to transfer all of these assets to the seller in order to maximize the surplus and the real
balance constraints are not binding. However, if there is not an abundance of assets, then buyers will not
be able to offer enough of the media of exchange to induce the seller to produce the optimal quantity. In
this case, since the size of the surplus is growing in q for all q < q ∗ , where q ∗ is the optimal quantity, the
buyer will spend everything he/she has to get the greatest surplus possible from trade.
Finally, note one additional result. As it becomes harder to verify whether or not an asset is counter-
feit, the fraction that is destroyed in the verification process becomes so large that the χi = 0 for asset i.
In this case, the corresponding real balance constraint is not binding for a trivial reason: the buyer will
never offer the particular asset to trade.
For now, let’s assume that λ1 > 1. Later, we will return to this assumption and figure out under what
conditions this is true. When λ > 1, equations (69) and (70) can be re-written, respectively, as
V (z, b) = γ[u(q) − c(q)] + W (z, b) (75)

36
 χz z + χb b = c(q) (76)
From the second equation, the implicit function theorem implies that

∂q χz
= 0 (77)
∂z c (q)

∂q χb
= 0 (78)
∂b c (q)
Differentiating V (z, b) with respect to z and b yields
 0 
0 0 ∂q u (q)
Vz = γ[u (q) − c (q)] + Wz = γχz 0 −1 +1 (79)
∂z c (q)
 0 
0 0 ∂q u (q)
Vb = γ[u (q) − c (q)] + Wb = γχb 0 −1 +1 (80)
∂b c (q)
Plugging these conditions back into the centralized market conditions yields
 0 
u (q)
ρ + π = αγχz 0 −1 (81)
c (q)
 0 
u (q)
ρ − r = αγχb 0 −1 (82)
c (q)
From these conditions, it must be true that
ρ+π χz
= (83)
ρ−r χb
Or,
(χz − χb )ρ − χb π
r= (84)
χz
The rate of return on bonds is therefore a function of the rate of time preference, the relative transaction
costs, and the rate of inflation. What this suggests is that money and bonds can have different real
rates of return when both circulate because of the differences in transactions costs. However, they are
related in a particular way. When the rate of inflation increases, individuals will want to economize on
money balances. They will sell money for bonds. This will drive down the rate of return on bonds. This
equation is essentially a no-arbitrage condition. If the real rate of return on bonds exceeds the right-hand
side of this equation, then everyone will prefer bonds to money. However, this drives down the rate of
return on bonds. This continues until the rate of return is equal to the right-hand side of the equation
above. To illustrate this idea consider the extremes. Suppose that there is no difference in verification
costs associated with money and bonds, i.e., χz = χb . Under this scenario, it is straightforward to see
that r = −π. In this case, the rates of return on money and bonds are equal. If transactions costs are
the same, then people should be indifferent between holding them – they are perfect substitutes. The
only way that people will be indifferent is if the rates of return are equal. Arbitrage will ensure this is
so. Similarly, consider the opposite extreme case. Suppose that money is perfectly recognizable, but the
entire bond is destroyed in the verification process. In that scenario, χz = 1 and χb = 0. Substituting
this into our equilibrium condition, we get r = ρ. So bonds cannot be used in transactions. Since they
cannot be used in transactions, the rate of return will have nothing to do with the rate of return on
money because they are no longer substitutes.

37
11.3 Implications for Monetary Aggregates
Our previous section suggests that it is possible for there to be different types of assets that circulate and
serve as a medium of exchange. Recall that when we discussed monetary aggregation in our previous
notes we said that simple sum aggregation is only appropriate if all media of exchange are perfect
substitutes. We can now interpret this result in the context of our model. Note that χi measures the
fraction of asset i that can be used in transactions in our model. This bears some resemblance to
Friedman and Schwartz’s idea of “money-ness.” Of course, in reality, we cannot observe χi directly.
Nonetheless, our model points us to something we can observe. Our model suggests that the relative
degree of “money-less” shows up in the relative rates of return of the assets. So how might we translate
this directly into the aggregation procedure? One way is through the use of Divisia aggregates. We
showed in the previous lecture how the weights of each component of the monetary aggregate are a
function of the relative rate of return between a benchmark rate and the rate of return of the asset.
An alternative measure that is a bit more straightforward to derive is what is know as the Currency
Equivalent Index, which was proposed by Rotemberg, Driscoll, and Poterba (1995). I will derive this here
because it is closely related to the implications of our model above.
Suppose that individuals have lifetime expected utility given as
∞
X
E0 β t u(ct , Lt ) (85)
t=0

where β ∈ (0, 1) is the discount factor, c is consumption, L is liquidity services, and uc , uL , −ucc , −uLL >
0.18 Furthermore, suppose that liquidity services are a function of the different types of money the indi-
vidual is holding:
L = f (m0 , m1 , m2 , . . . , mn )
where mi , i = 0, . . . n, are n + 1 types of media of exchange and m0 is currency. We will assume that f
is linearly homogeneous so that it follows that

Lt = f0 m0 + f1 m1 + f2 m2 + . . .

where fi denotes the partial derivative of L with respect to asset i.

The budget constraint of the individual can be written as

πt (at + m0,t + m1,t + . . . ) = Ra,t−1 at + R0,t−1 m0,t−1 + R1,t−1 m1,t−1 + R2,t−1 m2,t−1 + · · · − ct (86)

where πt+1 is the inflation rate, Ra is the (gross) nominal return on illiquid asset a, Ri is the (gross)
nominal return on money asset i. By gross return, we mean that if the percentage rate of return is 5%,
then Ri = 1 + ri = 1.05.
The relevant Langrangian equation can be written

max L = β j u(ct+j , Lt+j ) + λt+j [Ra,t+j at + R0,t+j m0,t+j + R1,t+j m1,t+j +

ct+j ,at+j ,mi,t+j

R2,t+j m2,t+j + · · · − ct+j − πt+j+1 (at+j+1 + m0,t+j+1 + m1,t+j+1 + . . . )]

The first-order conditions are
β j uc (ct+j , Lt+j ) = λt+j (87)
λt+j Ra,t+j = λt+j−1 πt+j (88)
18
Yes, we are cheating a bit here because we are assuming that people get some satisfaction from liquidity services and not
money stocks directly; may my sins be forgiven.

38
 β j uL (ct+j , Lt+j )fi,t+j + Ri,t+j λt+j = λt+j−1 πt+j i = 0, 1, . . . , n (89)
These conditions hold for all j. Note that we can combine the second two conditions to get:

β j uL (ct+j , Lt+j )fi,t+j + Ri,t+j λt+j = λt+j Ra,t+j (90)

Re-arranging this yields:

β j uL (ct+j , Lt+j )fi,t+j = λt+j (Ra,t+j − Ri,t+j ) (91)

This equation holds for all i and all j. Now, let’s compare asset any asset i = 1, 2, . . . , n to asset i = 0.
We can do this by dividing the condition above by the corresponding condition for asset i = 0. In other
words,
β j uL (ct+j , Lt+j )fi,t+j λt+j (Ra,t+j − Ri,t+j )
j
= (92)
β uL (ct+j , Lt+j )f0,t+j λt+j (Ra,t+j − R0,t+j )
which reduces to
fi,t+j (Ra,t+j − Ri,t+j )
= (93)
f0,t+j (Ra,t+j − R0,t+j )
Now we have an expression of the relative weight of all assets in comparison to asset i = 0. Let’s
assume that m0,t is currency and let’s assume that currency is perfectly liquid in that an additional unit
of currency provides an additional unit of liquidity services, f0,t = 1. Furthermore, note that currency
does not pay any nominal interest. Thus, R0 = 1 + r0 = 1. This implies that
1 + ra,t+j − 1 − ri,t+j ) ra,t+j − ri,t+j
fi,t+j = = (94)
1 + ra,t+j − 1 ra,t+j
Recall that we assumed that Lt was linearly homogeneous. Plugging this into Lt yields:
n  
X ra,t − ri,t
Lt = m0,t + mi,t (95)
ra,t
i=1

This implies that our measure of liquidity services is just a weighted sum of each individual asset that
can be spent as money. Furthermore, since we made the normalization that f0 = 1, we can see from the
weights that all other assets, fi ∈ [0, 1]. Let’s think about this in terms of our model from the previous
section. Suppose that χz = 1. It follows that χb has the interpretation of the relative degree of liquidity
of asset b (the bond). Recall that if χz = 1, then

r = (1 − χb )ρ − χb π

This is the real rate of return on the bond. This implies that the nominal rate of return on the bond is:

i = (1 − χb )ρ − χb π + π = (1 − χb )(ρ + π)

If the asset is just as liquid as currency, then χb = 1. This implies that the completely iliquid asset return
as ρ + π. Then substitute this into the formula we have from Rotemberg, et al. and we get
ρ + π − (1 − χb )(ρ + π)
fb = = χb
ρ+π
Thus, our search model above implies that the proper way to measure the aggregate quantity of money
is by using the Currency Equivalent Index.19
19
This is also consistent with the previous ideas of Hutt (1963) and Kessel and Alchian (1962).

39
References
[1] Andolfatto, David. 2015. “A Model of U.S. Monetary Policy Before and After the Great Recession.”
Federal Reserve Bank of St. Louis Review, Third Quarter, p. 233 - 256.

[2] Barnett, William A. 1978. “The User Cost of Money.” Economics Letters, Vol. 1, No. 2, p. 145 - 149.

[3] Barnett, William A. 1980. “Economic Monetary Aggregates: An Application of Index Number and
Aggregation Theory.” Journal of Econometrics, Vol. 14, p. 11 - 48.

[4] Belongia, Michael T. 1996. “Measurement Matters: Recent Results from Monetary Economics Re-
visited.” Journal of Political Economy, Vol. 104, No. 5, p. 1065 - 1083.

[5] Champ, Bruce, Scott Freeman, and Joseph Haslag. 2011. Modeling Monetary Economies, 3rd Edition.
Cambridge: Cambridge University Press.

[6] Cole, Harold L. and Lee E. Ohanian. 2004. “New Deal Policies and the Persistence of the Great
Depression: A General Equilibrium Analysis.” Journal of Political Economy, Vol. 112, No. 4, p. 779 -
816.

[7] Davidson, Paul. 2012. John Maynard Keynes. London: Palgrave.

[8] Fremling, Gertrud M. 1985. “Did the United States Transmit the Great Depression to the Rest of the
World?” American Economic Review, Vol. 75, No. 5, p. 1181 - 1185.

[9] Friedman, Milton. 1969. “The Optimum Quantity of Money,” in The Optimum Quantity of Money and
Other Essays. Chicago: University of Chicago Press.

[10] Friedman, Milton. 1989. “The Quantity Theory of Money,” in John Eatwell, Murray Milgate, and
Peter Newman (eds.), The New Palgrave: Money. New York: W.W. Norton.

[11] Friedman, Milton. 2003. “The Fed’s Thermostat.” Wall Street Journal, August 19, 2003.

[12] Friedman, Milton and Anna J. Schwartz. 1963. “Money and Business Cycles.” Review of Economics
and Statistics, Vol. 45, No. 1. Reprinted in Milton Friedman, The Optimum Quantity of Money and
Other Essays. Chicago: University of Chicago Press.

[13] Glasner, David. 1989a. Free Banking and Monetary Reform. Cambridge: Cambridge University Press.

[14] Glasner, David. 1989b. “On Some Classical Monetary Controversies.” History of Political Economy,
Vol. 21, No. 2, p. 201 - 229.

[15] Hendrickson, Joshua R. 2014. “Redundancy or Mismeasurement? A Reappraisal of Money.” Macroe-

conomic Dynamics, Vol. 18, p. 1437 - 1465.

[16] Hendrickson, Joshua R. 2015. “Monetary equilibrium.” Review of Austrian Economics, Vol. 28, No. 1,
p. 53 - 73.

[17] Hendrickson, Joshua R. 2017. “An Evaluation of Friedman’s Monetary Instability Hypothesis.” South-
ern Economic Journal, Vol. 83, No. 3, p. 744 - 755.

[18] Hicks, John R. 1936. “A Suggestion for Simplifying the Theory of Money.” Economica, Vol. 2, No. 5,
p. 1 - 19.

40
[19] Humphrey, Thomas M. 1981. “Adam Smith and the Monetary Approach to the Balance of Payments.”
Federal Reserve Bank of Richmond Economic Review, November/December, p. 3 - 10.

[20] Hutt, W.H. 1963. Keynesianism: Retrospect and Prospect. Chicago: Henry Regnery.

[21] Johnson, Noel and Mark Koyama. 2017. “States and Economic Growth: Capacity and Constraints.”
Explorations in Economic History, Vol. 64, No. 2, p. 1 - 20.

[22] Kessel, Reuben A. and Armen A. Alchian. 1962. “Effects of Inflation.” Journal of Political Economy,
Vol. 70, No. 6, p. 521 - 537.

[23] Kiyotaki, Nobuhiro and John Moore. 2002. “Evil is the Root of All Money.” American Economic
Review, Vol. 92, No. 2, p. 62 - 66.

[24] Lagos, Ricardo and Randall Wright. 2005. “A Unified Framework for Monetary Theory and Policy
Analysis.” Journal of Political Economy, Vol. 113, No. 3, p. 463 - 484.

[25] Laidler, David. 1981. “Adam Smith as a Monetary Economist.” Canadian Journal of Economics, Vol.
14, No. 2, p. 185 - 200.

[26] Laidler, David. 1999. Fabricating the Keynesian Revolution. Cambridge: Cambridge University Press.

[27] Leijonhufvud, Axel. 1968. On Keynesian Economics and the Economics of Keynes.

[28] McCloskey, Donald N. and J. Richard Zecher. 1976. “How the Gold Standard Worked, 1880 - 1913”
in Frenkel and Johnson (eds.), The Monetary Approach to the Balances of Payments. Toronto: University
of Toronto Press.

[29] Menger, Carl. 1892. “On the Origin of Money.” Economic Journal, Vol. 2, p. 239 - 255.

[30] Niehans, Jurg. 1979. The Theory of Money. Baltimore: Johns Hopkins University Press.

[31] Nosal, Ed and Guillaume Rocheteau. 2011. Money, Payments, and Liquidity. Cambridge: MIT Press.

[32] Ostroy, Joseph M. 1973. “The Informational Efficiency of Monetary Exchange.” American Economic
Review, Vol. 63, No. 4, p. 597 - 610.

[33] Rotemberg, Julio J., John C. Driscoll, and James M. Poterba. 1995. “Money, Output, and Prices:
Evidence from a New Monetary Aggregate.” Journal of Business and Economic Statistics, Vol. 13, No.
1, p. 67 - 83

[34] Rowe, Nick. 2010. “Milton Friedman’s Thermostat” Worthwhile Canadian Initiative, Decem-
ber 22, 2010. http://worthwhile.typepad.com/worthwhile_canadian_initi/2010/12/milton-friedmans-
thermostat.html

[35] Selgin, George and Lawrence H. White. 1987. “The Evolution of a Free Banking System.” Economic
Inquiry, Vol. 25, No. 3. Reprinted in George Selgin, Bank Deregulation and Monetary Order. New York:
Routledge, 1996.

[36] Serletis, Apostolos. 2007. The Demand for Money, Second Edition. New York: Springer.

[37] Smith, Adam. 1776. “Of the Origin and Use of Money,” in An Inquiry Into the Nature and Causes of
the Wealth of Nations, Book 1, Ch. 4.

[38] Sumner, Scott. 2015. The Midas Paradox. Oakland: Independent Institute.

41
[39] Thompson, Earl. 1974. “The Theory of Money and Income Consistent with Orthodox Value Theory,”
in Horwich and Samuelson (eds.), Trade, Stability, and Macroeconomics: Essays in Honor of Lloyd
Metzler.

[40] Thompson, Earl. 1997. “The Gold Standard: Causes and Consequences,” in Glasner (ed.), Business
Cycles and Depressions: An Encyclopedia. Routledge.

[41] Thompson, Earl and Charles Hickson. 2001. Ideology and the Evolution of Vital Economic Institu-
tions: Guilds, the Gold Standard, and Modern International Cooperation. Boston: Kluwer Academic
Publishers.

[42] Tobin, James. 1961. “Money, Capital, and Other Stores of Value.” American Economic Review, Vol. 51,
No. 2, p. 26 - 37.

[43] Tobin, James. 1980. “Discussion.” in John H. Kareken and Neil Wallace (eds.), Models of Monetary
Economies. Minneapolis: Federal Reserve Bank of Minneapolis.

42
 Asset Pricing Notes

Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
At the end of our consumption notes, we discussed the consumption and portfolio decisions of individuals
given that the pricing process for the asset was known. In what follows, we will be doing the opposite.
We will be determining the asset price (or asset prices) or expected returns using a model.
We will begin by using a version of Derman’s (2002) model, which effectively replicates the Arbitrage
Pricing Theory of Stephen Ross (1976). The basic idea of the model is to assume that there are no
opportunities for risk-free arbitrage. We then conjecture time series processes for asset prices and then
construct portfolios that are sufficiently large to diversify away the idiosyncratic risk of stocks. The only
source of risk remaining is aggregate risk. We then show how we can construct a larger portfolio that
consists of smaller sub-portfolios of assets – our original portfolio and (possibly various) portfolios that
track risk aggregate risk factors – to eliminate aggregate risk. The result is a model that explains the
excess expected return on any given portfolio as a linear function of the expected excess returns on
portfolios that track risk factors. Models like the CAPM model and various multi-factor models from
empirical finance can be thought of as special cases of this approach.
After this basic arbitrage argument, we introduce Lucas model of asset pricing. This model is a
consumption-based capital asset pricing model. The important thing about this model is that it connects
asset pricing with the consumption frameworks that we used in the previous lectures. We will show how
this model gives us a similar result to our arbitrage model.
We will then discuss the pricing of financial options. In particular, we will discuss the fact that option
pricing has important implications beyond financial options. In particular, we will discuss how the basic
structure of financial options can be used to introduce the concept of real options for investment.
Finally, I conclude with a discussion of the Efficient Markets Hypothesis and misconceptions there-
about.

2 Arbitrage Pricing
If you talk to enough people about asset pricing, you will begin to hear people refer to an asset’s
“fundamental value.” But what does this mean? Set aside assets for a moment and think about goods.
What is the fundamental value of bread? How would you answer this question? If someone tells you that
bread is expensive, what does this mean? The simplest answer is that the price of bread is high relative to
other goods and services that you could consume. But how do you know that the price is too high? And
is the price too high for everyone? Willingness to pay is a function of income, prices of related goods,
and preferences. The value of bread to you depends on your marginal utility. The equilibrium value of
bread in the market depends not only on your marginal utility, but the marginal utility of others. To
complicate matters, how you value bread, or any other good, is different from how you value a financial
asset. Consuming bread gives you some direct marginal utility. The utility that you get from owning

1
an asset is indirect and comes in the form of future consumption. As a result, we can think about asset
prices in the context of consumption decisions. For example, how much should I pay for this asset given
the way in which I want my consumption to evolve over time? Answering this question ties the asset price
to a concept of marginal utility. Alternatively, consider that when we think the price of bread is high
relative to the prices of other goods, we decide to buy less bread. But we don’t necessarily know why the
price of bread went up and none of us are consciously writing down and solving a utility maximization
problem in the aisle of the grocery store. In other words, we are making an assessment based on relative
valuation. Given the price of the goods, the price of bread seems high. We could do the same thing with
asset prices.
This idea can help us to think about two different ways of pricing financial assets. The first way to
price financial assets is do so based on relative valuation. For example, if I know the price of asset X,
then how much should I pay for asset Y ? We pursue this line of thinking in this section by starting
with a simple assumption. We assume that there are no risk-free arbitrage opportunities. If there were
risk-free arbitrage opportunities, then investors would try to capitalize on these opportunities. In doing
so, the very process of arbitrage would serve to eliminate the profit opportunity. I will show that this
line of thinking provides important implications for expected returns on individual assets and portfolios.
Namely, this approach suggests that the expected rate of return on any asset is equal to the risk-free rate
of return plus an idiosyncratic risk premium and a market risk premium. If you can construct portfolios
that are sufficiently large, then you can diversify away all of the idiosyncratic risk. Our model is therefore
consistent with Stephen Ross’s (1976) Arbitrage Price Theory. However, we present the argument in a
slightly different way. It also helps to provide some theoretical underpinnings associated with the various
factor models used in empirical finance.
The second way to price assets is to use a framework that incorporates the fact that asset purchases
today help to finance future consumption. Thus, the question is how much one should be willing to
pay for an asset given the correlation (if any) between the asset’s rate of return and the individual’s
consumption. We will leave this for the next section.
It is important to note that throughout our derivations of the various asset pricing models we will
never again use the term “fundamental value.” I say this because the idea of fundamental value is a
theoretical construct in the sense that we can take a theoretical model in which we are assumed to know
the fundamental value of the asset and we introduce various frictions into the model to determine whether
traders will ever end up discovering the fundamental value or whether prices can deviate significantly
from fundamental value (a so-called “bubble”). In other words, the “fundamental value” in these models
should really be called the “frictionless price,” with the analysis meant to identify what sorts of frictions
prevent us from getting to that price. But while “fundamental value” is a useful theoretical construct, it is
a meaningless empirical concept. We can never truly know the fundamental value of an asset any more
than we can know the fundamental value of bread. In fact, the very nature of subjective valuation that
has been at the core of economics since the marginal revolution implies that there is no such thing as
fundamental value. This is an important point to consider when discussing financial market efficiency
and the Efficient Markets Hypothesis, which we will return to at the end of the lecture.

2.1 A Derivation
Let’s assume that there are i = 1, . . . , N stocks. Further, let the price of stock i be written as Si and
conjecture that in equilibrium,
dSi
= µi dt + σi dzi (1)
Si
where µi is the expected return,
√ σi is the conditional standard deviation, and dzi is an increment of a
Wiener process (i.e. dzi = i dt, where i is drawn from a standard normal distribution). We will denote

2
the correlation between returns on stocks i and j as dzi dzj = ρij dt.
Now let’s begin with a basic idea: we will assume that there are no risk-free arbitrage opportunities
in equilibrium. In other words, it is not possible for me to profit in a financial market unless I am willing
to take on some degree of risk. A different way of saying this that will be helpful going forward is that:
• Two portfolios with the same irreducible risk will have the same expected return.
We can get a lot of mileage out of this idea. In what follows, we will consider markets under four different
scenarios:
1. Stocks are uncorrelated and not diversifiable.
2. Stocks are not diversifiable, but are correlated with the market.
3. Stocks are uncorrelated in a market that is diversifiable.
4. Stocks are diversifiable and correlated with the market.

2.1.1 Uncorrelated Stocks in an Undiversifiable Market

Suppose there are three assets, two stocks and a risk-free bond. Let’s conjecture that the prices of each
stock follow a geometric Brownian motion in equilibrium:
dS1
= µ1 dt + σ1 dz1 (2)
S1
dS2
= µ2 dt + σ2 dz2 (3)
S2
where we assume that E(dz1 dz2 ) = 0. In other words, the stocks are uncorrelated.
Let the risk-free rate of return on the bond be r, such that
dB
= rdt (4)
B
Given these 3 assets, how should an individual allocate his or her portfolio? And what are the implications
for asset prices?
We have conjecture that the individual buying stock 1 gets an expected rate of return µ1 whereas an
individual buying stock 2 gets an expected rate of return µ2 . However, these are conjectures about what
is true in equilibrium. How do we know what the expected return is on any given stock? Well, let’s use an
example that fits with our assumption. Suppose that I put together a portfolio of stock 2 and the risk-free
bond. We have assumed above that, in equilibrium, there should not be any opportunities for risk-free
arbitrage. In the absence of risk-free arbitrage, if I construct my portfolio to have the same risk as stock
1, then my portfolio should also have the same expected rate of return as stock 1. Let’s explore this idea
further. For example, let w denote the fraction of the portfolio that is devoted toward stock 2. The value
of the portfolio is given as
V = wS2 + (1 − w)B
Assume that w is not changing over time. Differentiating this yields

dV = wdS2 + (1 − w)dB

This implies that

 
dV 1
= w(µ2 S2 dt + σ2 S2 dz2 ) + (1 − w)rBdt (5)
V wS2 + (1 − w)B

3
Or
dV
= µV dt + σV dz2 (6)
V
where
wµ2 S2 + (1 − w)rB
µV =
wS2 + (1 − w)
wσ2 S2
σV =
wS2 + (1 − w)B
Note that it is possible to choose w such that the portfolio has the same volatility as stock 1. Specifically,
choose w such that
wσ2 S2
σ1 =
wS2 + (1 − w)B
Solving for w,
σ1 B
w= (7)
[(σ2 − σ1 )S2 + σ1 B]
Now, let’s consider the basic principle that we outlined at the beginning of this section. There should be
no opportunities for risk-free arbitrage. Since we have chosen w so that the risk of our portfolio is the
same as stock 1, then consider what would happen if the expected rates of return were different. If my
portfolio has a higher expected return than stock 1, but the same risk, then I can get a higher expected
rate of return without taking on any more risk. However, if this is true, then everyone will want to
sell stock 1 and buy the portfolio. However, by doing so, this alters the relative rates of return on the
assets. The assumption that there is no risk-free arbitrage therefore implies that with the same amount
of risk-taking, my portfolio and stock 1 should have the same expected rate of return. From the definition
of µV above, this implies
wµ2 S2 + (1 − w)rB
µ1 =
wS2 + (1 − w)B
Solve this equation for 1/w to get:
1 (µ2 − µ1 )S2
=1+ (8)
w (µ2 − r)B
Note that this solution for w has to be equal to the solution for w in equation (7), by definition. Re-writing
equation (7) yields
1 (σ2 − σ1 )S2
=1+
w σ1 B
Combining these equations and using some algebraic manipulation we get:
µ1 − r µ2 − r
= (9)
σ1 σ2
These are Sharpe ratios! This equality says that, with arbitrage pricing, the Sharpe ratio should be iden-
tical for all stocks. In other words, the expected excess rate of return on a stock should be proportional
to its risk. In fact, let λ be the Sharpe ratio. This equation implies that

µi = r + λσi

for all stocks. Thus, we can think of λ as the “price of risk.” It follows that the expected rate of return on
the stock is equal to the risk-free rate of return plus the product of the price of risk and the magnitude
of risk.

4
2.1.2 Undiversifiable Stocks that are Correlated with the Market
In the first example, we started with two stocks that each had idiosyncratic risk and a riskless bond. In
that example, there was no aggregate risk. In this example, we will assume that stocks are correlated with
the market. In other words, we will assume that aggregate news affects all stocks. For example, suppose
that the Bureau of Labor Statistics releases data on the labor market. This provides information about
the state of the economy and individual stock prices adjust to this news. Some stocks will tend to move
in the same direction as the market. Others will move in the opposite direction. So consider that you are
an investor and you want to try to eliminate aggregate risk. One way to judge how sensitive this stock
is to aggregate risk is to look at the covariance between the stock you own and the market as a whole.
You are interested in both the sign and the magnitude. If the covariance is negative then your stock
tends to go up when there is bad news and the rest of the market is going down. If the absolute value
of the covariance is large, then the stock is very sensitive to aggregate shocks. Ideally, we would like to
eliminate aggregate risk. Since the covariance between the stock and the market gives us an idea about
the sensitivity of the stock to aggregate shocks, we can use this to our advantage. Allow me to explain.
Suppose that you could construct a portfolio that tracks the overall market (or, these days, you could
just trade an exchange traded fund (ETF) or an index fund). Well, then I could construct a portfolio that
consists of my stock and my market portfolio. I can then choose the relative weights of these components
to eliminate aggregate risk. We will follow this logic to its conclusion and see what this tells us about
asset prices.
Well, let’s start by thinking about a simple portfolio in which we own one share of stock i and n
shares of the market:
V = Si + nM (10)
where M is the price of the market and Si is the price of stock i. We want to choose n to eliminate all
of the market risk.
To derive the equilibrium expected return on stock i, let’s first conjecture that value of the stock
market index follows a geometric Brownian motion in equilibrium such that
dM
= µM dt + σM dzM (11)
M
There is a risk-free asset, a discount bond, such that
dB
= rdt (12)
B
Finally, let’s continue to conjecture that equilibrium stock prices can be written:
dSi
= µi dt + σi dzi (13)
Si
where q
dzi = ρi,M dzM + (1 − ρ2i,M )dω
where dω is an increment of a Wiener process. Note here that there are two sources of risk when we buy
the stock, market risk and idiosyncratic risk.
The change in the value of our portfolio over time is
dV = dSi + ndM
Plugging in our conjectured solutions, we can re-write this as
q
dV = µi Si dt + σi Si ρi,M dzM + σi Si (1 − ρ2i,m )dωz + nµM M dt + nσM M dzM (14)

5
Or, simplifying,
q
dV = [µi Si + nµM M ]dt + [σi Si ρi,M + nσM M ]dzM + σi Si (1 − ρ2i,m )dω (15)

Now we see that the change in the value of the portfolio is subject to the idiosyncratic risk of the particular
stock and the aggregate, or market risk. As I stated above, what we want to do is choose n to try to
eliminate market risk. We can eliminate market risk by setting:

σi Si ρi,M + nσM M = 0

Solving this equation for n tells us the number of shares we need to sell short to eliminate market risk:
ρi,M σi Si
n=− (16)
σM M
Note that this implies that if the stock is positively correlated with the market, then I want to short the
market. However, if the stock is negatively correlated with the market then I want to buy the market.
Now, let’s take this expression and multiply both sides by σM to get
ρi,M σi σM Si
n=− 2 (17)
σM M

Why did we do this? Well, recall that a correlation coefficient, ρi,M has the following definition:

cov(Si , M )
ρi,M =
σi σM
Substituting this into the condition for n yields

cov(Si , M ) Si Si
n=− 2 = −βi,M (18)
σM M M

where β is the slope coefficient of a linear regression of Si on M . Plugging this into equation (10) gives
us
Si V
V = Si − βi,M M = (1 − βi,M )Si =⇒ Si = (19)
M 1 − βi,M
Furthermore, from equation (15) our choice of n implies that the change in our portfolio is given as
q
dV = [µi − βi,M µM ]Si dt + σi Si (1 − ρ2i,m )dω (20)

Or, using equation (19), we have

q
dV [µi − βi,M µM ] σi (1 − ρ2i,m )
= dt + dω (21)
V 1 − βi,M 1 − βi,M

Define µV and σV as
µi − βi,M µM
µV = (22)
1 − βi,M
q
σi (1 − ρ2i,m )
σV = (23)
1 − βi,M

6
We know from our arbitrage conditions in the first example that in equilibrium there is no risk-free
arbitrage and therefore,
µV − r = λσV (24)
Substituting our solution for µV and σV into this condition gives us
q
µi − βi,M µM σ i (1 − ρ2i,m )
−r =λ (25)
1 − βi,M 1 − βi,M

Multiplying both sides of (1 − βi,M ) and re-arranging yields:

q
µi − r = βi,M (µM − r) + λσi (1 − ρ2i,m ) (26)

This suggests that the expected excess rate of return on the asset is a function of the co-movement of the
asset with the market, the market’s excess return, and the idiosyncratic risk of the asset.

2.1.3 Uncorrelated Stocks in a Market Where Diversification is Possible

If there are a large number of stocks, we could potentially create a huge portfolio so that the idiosyncratic
risk of each individual stock doesn’t matter for our portfolio. In this section, we consider what would
happen if there was some large number of stocks, N , such that we can diversify away the idiosyncratic
risk.
Suppose there are N stocks that are uncorrelated. We could construct a portfolio of stocks such that
the value, V , of the portfolio is given as
XN
V = li Si (27)
n=1

where li is the number of shares of stock i and Si is the price of stock i. Then the evolution of the
portfolio is
N
X N
X N
X N
X
dV = li dSi = li (µi Si dt + σi Si dzi ) = li Si µi dt + li Si σi dzi (28)
n=1 n=1 n=1 n=1

It follows that we can write

N N
dV X X
= wi µi dt + wi σi dzi (29)
V
i=1 i=1

The expected return on the portfolio is

  N
dV /V X
E = µV = wi µ i (30)
dt
i=1

The variance is
N
X N
X X
σV2 = wi wj ρij σi σj = wi2 σi2 + wi wj ρij σi σj (31)
i,j=1 i=1 i6=j

where ρij is the correlation between stocks i and j. If N is large and the stocks are uncorrelated, then

σV2 ≈ 0

7
But think about our arbitrage condition above. We said that µi − r = λσi . If there is no variance in our
portfolio then
µV = r (32)
If we can diversify away all of the risk associated with stocks then it follows that this portfolio must earn
the risk-free rate of return. Otherwise, this would violate the idea that there are no risk-free arbitrage
opportunities.

2.1.4 Diversifiable Stocks Correlated with the Market

Let’s imagine that instead of buying one stock and a market portfolio/ETF, we construct our desired
portfolio and then we buy/sell the market ETF to remove all of the aggregate risk. Let µP denote the
expected return of the portfolio and σP2 denote the variance of the portfolio. It follows that
q
µP − r = βP,M (µM − r) + λσP (1 − ρ2P,m ) (33)

If the riskiness of each individual stock is idiosyncratic and diversifiable, then if the investor builds a
large enough portfolio, we would have the result that σP ≈ 0. Thus, our no-arbitrage equilibrium would
imply that
µP = r + βP,M (µM − r) (34)
This is the result of the Capital Asset Pricing Model! It says that the excess rate of return on our portfolio
is equal to β times the excess return of the stock market, where β measures the sensitivity of the portfolio
to changes in the overall market.

2.2 Implications
Our simple arbitrage model implies the following:

• The Sharpe ratio applies to all assets. In other words, assets with higher risk require higher
expected returns.

• If one can construct a large enough portfolio of assets with idiosyncratic risk, then one can suffi-
ciently diversify to eliminate this risk.

• Assets tend to move together. This is aggregate, or market, risk. Assets that have greater co-
movement with the overall market have greater risk and therefore must have a higher expected
return.

• If one can construct a large enough portfolio to eliminate the idiosyncratic risk of the assets, then
we arrive at the Capital Asset Pricing Model that says that expected excess returns are only a
function of the market’s excess return and the asset’s co-movement with the market.

• Our results are therefore consistent with Ross’s one-factor model of arbitrage pricing.

• Our model required no assumptions about utility maximization and the functional form of the
utility function. In fact, we didn’t model decision-making at all other than to argue that people
seek to capitalize on risk-free arbitrage to the extent that there are no such opportunities in
equilibrium.

8
2.3 Multifactor Models: What Are They and How Can We Reconcile Them With Our
Pedagogical Example?
The Capital Asset Pricing Model has been found to be lacking in empirical performance. Allow me to
explain what this means. Suppose that you wanted to test the CAPM. What you could do is estimate a
regression of the form:
(ri,t − rt ) = α + β(rM,t − rt ) + et (35)
where ri,t is the observed rate of return on asset/portfolio i at time t, rt is the risk-free rate of return at
time t, and rM,t is the observed rate of return on the market at time t. If the CAPM explains everything
there is to know about excess returns, then α = 0. This is a testable hypothesis. If you estimated this
regression and you can reject the null hypothesis that α = 0, then this implies that there is some excess
return that the asset/portfolio generates that is not explained by market risk. What I mean when I say
that the CAPM is lacking empirically is that α 6= 0 in these sorts of regressions.
So how can we reconcile this with our model above? Or, put differently, how can I justify spending
so much time deriving something that does not fit the data particularly well. There are two possible
answers to this question. First, recall that in the example we used to illustrate the CAPM result, we
assumed that the stock was subject to both idiosyncratic risk and aggregate risk. The only way that the
regression shown in equation (35) would fit the data well is if (a) we can sufficiently diversify away all of
the idiosyncratic risk, or (b) the sources of idiosyncratic risk are unobservable. In reality, it is likely that
neither of these conditions is satisfied. In other words, the vast majority of market participants are not
going to be able to create a portfolio large enough to diversify away all of the idiosyncratic risk of the
individual stocks in the portfolio. In addition, stocks (or, more appropriately, the underlying firms) might
have characteristics that help us to predict the amount of idiosyncratic risk that will be evident in the
stock.
Second, suppose that we can sufficiently diversify away the risk (for example, as individuals we are
unlikely to be able to do so, but large hedge funds and mutual funds might be able to pool resources
to do so). Then it is possible that aggregate risk is multifaceted. For example, maybe aggregate shocks
affect small firms more than large firms. If you know this, then you would want to hedge against this
risk as well. Using the same logic that we used to derive the CAPM result above, we could derive the
following condition for our portfolio:
µp = r + βp,M (µM − r) + βp,S (µS − r) (36)
where µS is the expected rate of return on a portfolio of “small” stocks. This is just Ross’s model for
two-factors. In fact, we can continue this process for as many factors as we like. This is important
because the empirical finance literature uses multifactor models. If you believe that the CAPM is the
correct model, then it is hard to justify why these factors should be included in an asset pricing model.
However, if you start from the premise of arbitrage pricing, then you realize that CAPM is actually just a
special case of Ross’s Arbitrage Price Theory. It follows that the three-factor model of Fama and French
(1993) that includes the market factor, a size factor, and a value factor is just one application of arbitrage
pricing theory. These types of factor models essentially assume that certain characteristics of a firm or a
stock lead to different responses to aggregate shocks. Others like Chen, Roll, and Ross (1986) argue that
there are specific macroeconomic factors that might affect stock returns in different ways. They suggest
5 different factors: industrial production, unexpected inflation, expected inflation, the term structure of
interest rates, and a risk premium (spread between low and high-rated bonds). Arbitrage Price Theory
again can be used to justify the Chen-Roll-Ross empirical approach. For example, we could construct
portfolios of assets that track each of these five indicators particularly well. If so, they a sufficiently
diversified portfolio would give us:
µp = r + βp,IP (µIP − r) + βp,U I (µU I − r) + βp,EI (µEI − r) + βp,T S (µT S − r) + βp,RP (µRP − r)

9
where the subscripts refer to portfolios that track each of the respective macroeconomic factors listed
above. This is just a five-factor version of Ross’s APT.

3 A Two-Period “Lucas” Asset Pricing Model

Consider an overlapping generations model in which an individual lives for two periods and has lifetime
utility given as
U = u(ct ) + βEt u(ct+1 ) (37)
where u0 , −u00 > 0. The individual receives an endowment, y, in each period. In the first period, the
individual can use their endowment to consume or purchase an asset for a price pt . In the second
period, the asset pays an uncertain dividend, dt+1 , and can be sold for a price, pt+1 . It follows that the
individual’s budget constraints when young and when old can be written, respectively, as

yt = ct + pt at (38)

yt+1 + (pt+1 + dt+1 )at = ct+1 (39)

The individual’s maximization problem can be written as

max u(yt − pt at ) + βEt u[yt+1 + (pt+1 + dt+1 )at ] (40)

at

The first-order condition for maximization is given as

− pt u0 (ct ) + βEt (pt+1 + dt+1 )u0 (ct+1 ) = 0 (41)

Or, solving this for pt yields,

u0 (ct+1 )
pt = βEt (pt+1 + dt+1 ) (42)
u0 (ct )
0
Defining mt+1 := β uu(c0 (ct+1
t)
)
as the discount factor and xt+1 := pt+1 + dt+1 , we can re-write this as

pt = Et mt+1 xt+1 (43)

where m is the time-varying discount factor and x is the payout.

4 Lucas’s Endowment Economy

Lucas’s (1978) model assumes that there are a large number of individuals, who have one type of asset:
trees. Unlike our previous example, we assume that the individuals (households) live forever. The total
number of trees is assumed to be fixed and equal to the number of individuals in the economy. Individuals
can buy and sell ownership shares in the trees. Let at denote the number of shares that an individual
owns at time t. These trees produce a dividend in the form of fruit. The amount of fruit that falls each
period is random. The fruit is not storable (just like real fruit, it doesn’t last forever). Let dt be the
random dividend such that the total fruit that the individual would receive at time t is dt at . Furthermore,
let pt be the price of an ownership share. The individual’s budget constraint can be written as

pt at+1 = (pt + dt )at − ct (44)

10
where at is the number of assets (“trees”) the individual owns and ct is consumption. Suppose that the
individual has expected lifetime utility of
∞
X
E0 β t u(ct ) (45)
t=0

where β ∈ (0, 1) is the discount factor and u0 , −u00 > 0.

The Bellman equation can be written as
 
pt + dt ct
V (at ) = max u(ct ) + βEt V at − (46)
ct ,at+1 pt pt
The first-order condition is
pt u0 (ct ) = βEt V 0 (at+1 ) (47)
Note that if we differentiate the value function, we have
pt + dt
V 0 (at ) = Et V 0 (at+1 ) = (pt + dt )u0 (ct )
pt
Iterating forward and substituting this into the first-order condition yields
pt u0 (ct ) = βEt u0 (ct+1 )(pt+1 + dt+1 ) (48)
Or, solving for pt yields:
u0 (ct+1 )
pt = βEt (pt+1 + dt+1 ) (49)
u0 (ct )
Note that this is identical to our two period model and can be defined in terms of equation (43). Re-write
the price equation as
pt u0 (ct ) = βEt u0 (ct+1 )(pt+1 + dt+1 ) = βEt u0 (ct+1 )pt+1 + βEt u0 (ct+1 )dt+1
Iterating forward yields
pt+1 u0 (ct+1 ) = βEt+1 u0 (ct+2 )pt+2 + βEt+1 u0 (ct+2 )dt+2
Substituting this into the original equation yields
pt u0 (ct ) = βEt u0 (ct+1 )dt+1 + β 2 Et [Et+1 u0 (ct+2 )pt+2 + Et+1 u0 (ct+2 )dt+2 ]
Using the law of iterated expectations yields:
pt u0 (ct ) = βEt u0 (ct+1 )dt+1 + β 2 Et [u0 (ct+2 )pt+2 + u0 (ct+2 )dt+2 ]
Continuing to do this over an infinite horizon yields1
∞  0 
j u (ct+j )
X
pt = Et β dt+j (50)
u0 (ct )
j=1

Thus, the price of the asset is equal to the discounted present value of all future dividends. Notice that
the discount rate is not necessarily constant over time.
1
Note that if we continue to iterate forward, we have:
s
X
pt u0 (ct ) = β s u0 (ct+s )pt+s + β j u0 (ct+j )
j=1

Since β < 1, taking the limit as s → ∞, this first term on the right-hand side goes to zero.

11
5 Consumption CAPM
The model that we discussed can be interpreted as a consumption-based Capital Asset Pricing Model
because it suggests that the relevant factor that the individual should care about is the time path of
consumption. For example, it is possible to write down a model consistent with our arbitrage approach
above to get a factor model of the form: µp = r + β(µc − r), where µp is the expected return on our
portfolio and µc − r is the expected excess rate of return on a portfolio that tracks consumption. In what
follows I will demonstrate how to generate some implications from the C-CAPM and discuss what has
come to be known as the Equity Premium Puzzle.
Let’s return to our consumption model. From our equilibrium conditions we have

u0 (ct+1 )
pt = βEt (pt+1 + dt+1 )
u0 (ct )

Dividing both sides by pt and defining the gross rate of return as Rt+1 = (pt+1 + dt+1 )/pt yields

u0 (ct+1 )
βEt Rt+1 = 1
u0 (ct )

Let’s take a first-order Taylor Series approximation of u0 (ct+1 ) around ct . This implies that

u0 (ct+1 ) = u0 (ct ) + u00 (ct )∆ct+1

Dividing both sides by u0 (ct ) yields

u0 (ct+1 ) u00 (ct )

= 1 + ∆ct+1
u0 (ct ) u0 (ct )
Now suppose that the utility function is
c1−σ
u(c) =
1−σ
Then,
u0 (c) = c−σ
u00 (c) = −σc−σ−1
and
u00 (c) σc−σ−1 σ
0
= − −σ
=−
u (c) c c
Plugging this into the Taylor series approximation yields:

u0 (ct+1 ) u00 (ct ) ∆ct+1

0
= 1 + 0
∆ct+1 = 1 − σ
u (ct ) u (ct ) ct
Now plug this into the original equilibrium condition to get:
 
∆ct+1
βEt 1 − σ Rt+1 = 1
ct

Recall that the gross rate of return is Rt = 1 + rt . This implies that

 
∆ct+1
βEt 1 − σ (1 + rt+1 ) = 1
ct

12
Carrying out the multiplication and using the fact that β = 1/(1 + ρ) yields
 
∆ct+1 ∆ct+1
1 − Et σ + Et rt+1 − σEt rt+1 = 1 + ρ
ct ct
Using the definition of covariance, we have
   
∆ct+1 ∆ct+1 ∆ct+1
ρ = Et rt+1 − σEt − σCov , rt+1 − σEt Et rt+1
ct ct ct
Solving for Et rt+1 , yields
 
ρ+ σEt ∆cct+1
t
+ σCov ∆ct+1
ct , rt+1
Et rt+1 = (51)
1 − σEt ∆cct+1
t

Note that the difference between a risk free asset and the risky asset is this covariance term. It follows
that the risk-free asset is given
f ρ + σEt ∆cct+1
t
rt =
1 − σEt ∆cct+1
t

It follows that the excess returns on the risky asset are given as
 
∆ct+1
Cov ct , rt+1
f
Et rt+1 − rt = σ (52)
1 − σEt ∆cct+1
t

Or, using the fact that for the risk-free asset,

 
∆ct+1
β 1 − σEt Et (1 + rtf ) = 1
ct
we can write  
∆ct+1
rtf
Et rt+1 − = σβ(1 + rtf )Cov , rt+1 (53)
ct
Thus, the C-CAPM implies that the risk of holding the risky asset depends on the covariance of con-
sumption growth and the rate of return. To make the intuition simple, consider that in a balanced growth
economy, the expected growth rate of consumption is equal to the expected growth rate of the economy.
The model therefore suggests that if the rate of return on the asset is higher when times are good (i.e.,
consumption growth is high), then this poses a risk to the owner of the asset. Why? Because this also
implies that when times are bad and consumption growth is low the rate of return on the asset is low as
well.
This seems reasonable enough. Nonetheless, we are interested in how well this model predicts the
data. One test of this would be to see if the model conforms well to standard assumptions in economics.
For example, we can measure consumption growth and the rate of return on risky assets like equity and
a risk-free asset like a government bond. We can then calculate the covariance between consumption
growth and equity returns and plug the mean values of consumption growth, and excess returns into our
equation. Then we can solve for σ, which measures the coefficient of relative risk aversion. If we carry
out this exercise, what we find is that the magnitude of σ is far too high than one would reasonably
expect from economic theory. This result is known as the equity premium puzzle because it suggests that
excess returns on equity are much too high to be consistent with standard assumptions of risk aversion.
There is an abundance of literature that tries to explain this puzzle. For an overview, see Mehra and
Prescott (2003). For a recent and interesting attempt to explain the equity premium puzzle by appealing
to the possibility of rare disasters with time-varying intensity, see Gabaix (2012).

13
6 Option Pricing
Financial options are unique financial contracts. For example, when you buy a stock your wealth fluctuates
with the price of the stock. However, a financial option is a contract that gives the purchaser the option
to buy or sell an asset at a given price at a particular date in the future.2 An option to buy the asset
at a particular price is a call option. An option to sell the asset at a particular price is a put option.
This option is valuable precisely because of the fact that it gives the owner an option to take action.
For example, suppose that I own a stock, but I am concerned that the stock will fall below a certain
price. I could buy a put option that gives me the option to sell the stock at a particular price, say K, at
a particular date in the future. I can therefore limit my losses in the event that the stock does decline.
However, even if the stock does not decline, I can decline the option to sell at the lower price.
Financial options are useful for a variety of reasons. First, they allow investors to bet on whether
a stock will move in a particular direction. Second, they allow people to hedge their long and short
positions in a stock to limit their losses. Third, for a given theory of option prices, we can use that
theory to gauge the risk the market sees for a particular stock (or portfolio of stocks). Of course, the
latter characteristic depends on our model.
A European call option on a stock with a price, S, with a strike price, K, that expires at time T then
has a value at expiration of
C(S, T ) = max{ST − K, 0} (54)
If this is the payout at time T , how can we price the contract at time t < T ? We can do this in one of
two ways, but both ways involve using the idea that there will be no opportunities for risk-free arbitrage
in the market. The first way that we can price the option is by supposing that you have a portfolio that
consists of an option and shares of the underlying stock. In that scenario, we note that it must be the
case that this portfolio provides a risk-free rate of return. The second way that we can price the option
is by suppose that you can construct a portfolio solely of options. In this case you could buy put and call
options on the same asset. You should not be able to earn profits from this approach. Thus, there should
be a put-call parity.
We will assume throughout that the stock price, S, follows a geometric Brownian motion:
dS
= µs dt + σs dz (55)
S

6.1 A Simple Pedagogical Model

6.1.1 The Absence of Risk-Free Arbitrage and Risk Neutral Pricing
Suppose that there are two assets, a bond and a stock. The world exists for two periods, today (time
t = 0) and tomorrow (time t = 1). The value of the bond evolves according to

B1 = (1 + r)B0

The interest rate is constant and risk-free. For simplicity, lets assume that B0 = 1. Then B1 = (1 + r).
The stock price evolves as follows,
S0 = s
S1 = sZ
2
In this section, we will deal with European-style option contracts. These are option contracts in which one can only exercise
the option at the expiration date. American-style options allow investors to exercise the option at any point in time prior to
the expiration date. This complicates matter significantly. We will not address these options now. However, when we discuss
capital investment in subsequent lectures, we will discuss how we can view capital investment decisions as akin to American
options with no expiration date.

14
where
u w/prob. pu

Z=
d w/prob. pd
where u > d and pu + pd = 1.3 Let x denote the quantity of bonds and y the shares of stock in one’s
portfolio. It follows that the value of the portfolio at times 0 and 1 are given as

V0 = xB0 + yS0

V1 = x(1 + r)B0 + yS1

From the definitions above, this can be written as

V0 = x + ys

V1 = x(1 + r) + ysZ

Definition 1. A risk-free self-financing portfolio satisfies the following two conditions:

V0 = 0

V1 > 0 w/prob. 1

In other words, a risk-free self-financing portfolio is a portfolio that requires no initial outlay by the
investor, but earns a positive rate of return with absolute certainty. Thus, the investor can earn a rate
of return without any risk. Now recall our assumption in our arbitrage pricing model above. In that
model our assumption is that there are no opportunities for risk-free arbitrage. Thus, we need to derive
a condition that rules out risk-free arbitrage.4 To do so, consider our portfolio definition implies that for
the first condition to hold, it must be the case that

x + ys = 0

Plugging this into V1 gives us

V1 = [Z − (1 + r)]ys
Or,
ys[u − (1 + r)] w/prob. pu

V1 =
ys[d − (1 + r)] w/prob. pd
If V1 > 0 with probability 1, then V1 > 0 in both scenarios. Since u > d by assumption, a sufficient
condition to guarantee that V1 is not positive with absolute certainty is if

ys[u − (1 + r)] > 0

ys[d − (1 + r)] < 0

Or,
u>1+r
d<1+r
3
This is a binomial model. I show in the appendix that Brownian motion has a binomial discrete-time representation.
4
Remember that our assumption of risk-free arbitrage just says that if there is a profit opportunity for anyone that does not
require risk, then individuals will seek to profit. As more and more people do this, then risk-free arbitrage opportunity will
vanish.

15
So if d < 1 + r < u, then there is no opportunity for risk-free arbitrage.
However, let’s carry this logic further. If 1 + r ∈ (d, u), then we can write
1 + r = qu u + qd d
where qu , qd ≥ 0 and
qu + qd = 1
Suppose that we define a probability measure Q such that Q(Z = u) = qu and Q(Z = d) = qd . We
can use this to derive an important result. Let E Q denote the mathematical expectation associated with
probability measure Q. Then, we can write
1 1 1 1
E Q (S1 ) = [qu us + qd ds] = s[qu u + qd d] = s(1 + r) = s = S0
1+r 1+r 1+r 1+r
In other words,
1
E Q (S1 )
S0 =
1+r
So today’s stock price is equal to the expected discounted value of next period’s stock price. This is just
the risk neutral valuation of the stock price. Thus, we can think of this Q measure as a risk neutral
probability measure. It also follows that an absence of risk-free arbitrage requires that a Q probability
measure exists.
Before introducing options into our model, we should note that we can solve explicitly for the risk-
neutral probabilities qu and qd . These probabilities satisfy the following two conditions:
qu u + qd d = (1 + r)
qu + qd = 1
Solving the section condition for qu and plugging into the first conditions yields:
qu (u − d) + d = (1 + r)
Or,
(1 + r) − d
qu =
u−d
It follows that
u − (1 + r)
qd =
u−d

6.1.2 Options Contracts

Now suppose that there is a financial derivative (a contingent claims contract) that has a payout based on
the price of the stock.5 Specifically, let’s suppose that the contingent claims contract gives an individual
the option, but not the obligation to buy the stock at price K at time t = 1. The payout of the contract
at t = 1 is assumed to be consistent with equation (54) such that
C(S, 1) = max{S1 − K, 0}
Or maintaining notation more consistent with this section:
C1 (S1 ) = max{S1 − K, 0}
The question that we want to answer is what we should pay for this contingent claim in period t = 0.
5
What follows is a two-period model based on Cox, Ross, and Rubinstein (1979). For a textbook treatment, see Bjork (2009).

16
Proposition 1. Suppose that there is no risk-free arbitrage. For some risk neutral probability measure, Q, the
price of the option today will be equal to the expected present discounted value of the option at maturity:
1
C0 = E Q (C1 )
1+r
Proof. Suppose that an individual constructed a portfolio of the risk-free bond and the stock such that
V1 = C1 . Since we have assumed that there are no opportunities for risk-free arbitrage, it must be the
case that V0 = C0 . So let’s consider a portfolio of stocks and bonds where B0 = 1 and S0 = s. It follows
that
V0 = x + ys
Now, assume that for some probability measure, P , the price of the stock in period 1 is S1 = sZ, where

u w/prob. pu

Z=
d w/prob. pd

If the value of the portfolio of bonds and stock is the same as the value of the option at time t = 1, then

(1 + r)x + yus = C1 (us) (56)

(1 + r)x + yds = C1 (ds) (57)

From the first equation,
C1 (us) − (1 + r)x
y=
us
Plugging this into the second equation:

C1 (us) − (1 + r)x
(1 + r)x + ds = C1 (ds)
us
Or,
1 uC1 (ds) − dC1 (us)
x=
1+r u−d
To get y, use equations (57) and equation (56) to write:

(1 + r)x + yus − (1 + r)x − yds = C1 (us) − C1 (ds)

Solving for y yields:

1 C1 (us) − C1 (ds)
y=
s u−d
Now, plug these solutions for x and y into V0 to get

1 uC1 (ds) − dC1 (us) C1 (us) − C1 (ds)

V0 = +
1+r u−d u−d
Or,
1 uC1 (ds) − dC1 (us) + (1 + r)C1 (us) − (1 + r)C1 (ds)
V0 =
1+r u−d
Collecting like terms and factoring:
 
1 (1 + r) − d u − (1 + r)
V0 = C1 (us) + C1 (ds)
1+r u−d u−d

17
Recall our definitions of qu and qd above. It follows that we can re-write
 
1
V0 = qu C1 (us) + qd C1 (ds)
1+r

Since V1 = C1 , it must be true that C0 = V0 . Otherwise, there would be an opportunity for risk-free
arbitrage. Thus,  
1 1
C0 = qu C1 (us) + qd C1 (ds) = E Q (C1 ) (58)
1+r 1+r

The implications of this approach are as follows:

• We can solve for the price of an option through backward induction. In other words, we can figure
out what a financial derivative is worth at expiration. Then one period before expiration, we know
that the value of the option is the expected present value of the option, where the expected value
is determined by using the risk neutral probability measure.

• The use of the risk neutral probability measure is due to the absence of risk-free arbitrage. Using
this probability measure DOES NOT imply that individuals are risk neutral. In fact, since the
valuation is derived from the “no risk-free arbitrage” assumption, this pricing measure does not
rely on assumptions about the preferences of those participating in the market.

We will now proceed to show how one can value options in continuous time. First, we will rely on the
principle of dynamic replication as we did in the example above using the Black-Scholes/Merton model.
Second, we will will use the principle of static replication by relying on the idea of put-call parity.

6.2 The Black-Scholes/Merton Model

Recall the option price as written in equation (54). Note that the price of an option contract can be
written as an implicit function of time and the price of the underlying asset. Thus, the price of a call
option at any time period t < T can be written C(S, t). Using Ito’s Lemma, it follows that

∂C ∂C 1 ∂2C
dC = dt + dS + (dS)2 (59)
∂t ∂S 2 ∂S 2
Given that we know dS, this implies

1 ∂2C 2 2
 
∂C ∂C ∂C
dC = + µs S + σ S dt + σs Sdz (60)
∂t ∂S 2 ∂S 2 s ∂S

Define
1 ∂2C 2 2
 
1 ∂C ∂C
µc := + µs S + σ S
C ∂t ∂S 2 ∂S 2 s
∂C S
σc := σs
∂S C
Now, we can write
dC
= µc dt + σc dz (61)
C
Suppose that I own this call option and I want to construct a portfolio such that I eliminate all the risk
of holding the option. What I could do is sell the stock short and use the revenue to buy the option. If

18
the stock price goes down, my call option is worthless, but I can buy back the stock that I borrowed to
short at a lower price and earn a profit. If the stock price goes up, I have to cover my short position. My
call option will allow me to earn a profit from the up-move in the stock that will eliminate what I need
to cover the short position. This sort of portfolio is possible, but the question is: how much of the stock
do I have to short to make this a risk-free portfolio?
Let V be the value of my portfolio such that V = nS + C, where I own one call option and n is the
number of shares of the stock that I am going to short. It follows that

dV = ndS + dC = nµs Sdt + nσs Sdz + µc Cdt + σc Cdz = (nµs S + µc C)dt + (nσs S + σc C)dz

My goal is to choose n to eliminate risk. Thus, I want to choose n such that

nσs S + σc C = 0

So,
σc C
n=−
σs S
Now recall that we assume that there is no risk-free arbitrage in this model. This implies that the
following must be true:
dV /V
=r
dt
where r is the risk-free rate of return. If this condition doesn’t hold then (a) there is a possibility of
earning a profit from this portfolio without taking on additional risk, or (b) the portfolio isn’t worth
constructing because one could do better by investing in the risk-free bond.
Note that if this condition holds, then

dV = rV dt = r(nS + C)dt

From the choice of n above, we must have

(nµs S + µc C)dt = r(nS + C)dt

Or, solving for n,

C µc − r
n=−
S µs − r
This n must be consistent with the choice of n that eliminates portfolio risk. So,
σc C C µc − r
=
σs S S µs − r
which simplifies to
µc − r µs − r
=
σc σs
Using the definitions of µc and σc , we get
 
1 ∂C ∂C 1 ∂2C 2 2
C ∂t + ∂S µs S + 2 ∂S 2 σ S −r
µs − r
∂C S
=
∂S σs C
σs

Multiplying the numerator and denominator by C and eliminating terms yields

1 ∂2C 2 2
 
∂C ∂C ∂C
+ µs S + 2
σ S − rC = (µs − r) S
∂t ∂S 2 ∂S ∂S

19
Simplifying
∂C ∂C 1 ∂2C
rC =
+ rS + σ2S 2 2 (62)
∂t ∂S 2 ∂S
This is the Black-Scholes equation, which has a solution of the form:
C(S, t) = St e−r(T −t) N (d1 ) − e−r(T −t) KN (d2 ) (63)
where  
er(T −t) S
ln + (σ 2 /2)(T − t)
K
d1 = √ (64)
σ T −t
 
r(T −t)
ln e K S − (σ 2 /2)(T − t)
d2 = √ (65)
σ T −t
and N (z) is the cumulative normal distribution:
Z z
1 2
N (z) = √ e−1/2y dy (66)
2 π −∞

6.3 Put-Call Parity

In the previous two sections, we showed how to price options using dynamic replication. Allow me to
explain what I mean. In the binomial model of Cox, Ross, and Rubinstein (1979), we start with a basic
assumption. Suppose that I can construct a portfolio that replicates the payout of the option contract at
expiration. It follows that the value of the option and the value of the portfolio must be equal through
time. Otherwise, I could earn profits through risk-free arbitrage. Some have expressed doubts about the
feasibility of the dynamic replication.6 In this section, I will show how to use static replication.
Consider a European call option and a European put option with the same strike price and same
expiration date. The value of the call and the put options, respectively, at expiration can be written as
C(S, T ) = max{ST − K, 0} (67)
P (S, T ) = max{K − ST , 0} (68)
Suppose that an investor bought the call option and sold the put option. The portfolio would be
C(S, t) − P (S, t). At the expiration date, the individual would get:
C(S, T ) − P (S, T ) = max{ST − K, 0} − max{K − ST , 0} = ST − K (69)
So, in other words, this portfolio gives me a guaranteed payout at the expiration date T . We can use this
relationship to price options. For example, suppose that there is a discount bond that pays a risk-free
rate of return, r. I could, in period t, sell short these discount bonds that will be worth K dollars at
time T and use the proceeds to buy the stock at price St . The value of this portfolio at time T would
therefore be the same as the portfolio of options, ST − K. The value of the portfolio at time t, however,
is St − e−r(T −t) K. This implies that:
C(S, t) − P (S, t) = St − e−r(T −t) K
So, if I know the price of the put, I can price the call option using:
C(S, t) = P (S, t) + St − e−r(T −t) K (70)
And, if I know the price of the call option, I can price the put option as
P (S, t) = C(S, t) − St + e−r(T −t) K (71)
6
For a discussion, see Derman and Taleb (2005).

20
6.4 Can Option Theory Be Applied to Other Types of Decision-Making?
Option contracts give individuals the option to buy or sell an asset at a pre-determined price at a
particular moment in time. The individual is under no obligation to exercise this option. If we think
about this basic idea more broadly, we can see that this idea can be related to other types of economic
decision-making. For example, in basic economic examples, we often suggest that people should engage
in activity in which the benefits exceed the costs. However, this is a static analysis. Sometimes, even if the
benefits currently exceed the costs, you might want to wait for more information. In this context, taking
action is similar to exercising an option. When you see that some type of activity has benefits that exceed
the costs, you have the option to engage in that activity or to wait for further information. To give an
example, suppose that you are a firm that is interested in entering a new product market. Entering this
market would entail a substantial upfront fixed cost. In addition, the demand for this product fluctuates
randomly over time. Given the substantial upfront cost and the fluctuations in demand, you might not
want to “exercise your option” to enter the market right now even if the current benefits exceed the costs.
You might prefer to wait for more information. Thus, having the option of whether or not to enter this
new product market has value. In fact, we can treat the option to enter a market as akin to owning a
financial option, in which profitability (which is a function of demand that is fluctuating randomly) is
the “price” and the upfront investment is the “strike price” on the option. Contrary to financial options,
however, these options do not necessarily have an expiration date.

7 The Efficient Markets Hypothesis

7.1 What It Is
The concept of the Efficient Markets Hypothesis (EMH) is perhaps most closely associated with Eugene
Fama. As an undergraduate, Fama had collected daily price data for one of his professors, Martin Ernst,
who published a newsletter on financial markets and hoped to use the data to identify trading strategies
that could be recommended to the newsletter readers. However, what Fama discovered was that none
of strategies that had worked in historical data remained successful strategies going forward. Given this
experience with the data, when he arrived at the University of Chicago for graduate school, Fama seemed
to find Harry Roberts’s random walk theory to be quite compelling (if we are to judge by Fama’s early
work). Roberts had argued that the charts of stock prices were indistinguishable from charts of a random
walk.7
Fama (1970: 383) formally defined the EMH as follows: “A market in which prices always ‘fully reflect’
available information is called ‘efficient.”’ A hypothesis requires testing. However, this definition of the
EMH does not lend itself well to testing. Fama (1970: 384) himself noted in the same paper that this
definition “is so general that it has no empirically testable implications.” As a result, Fama proposed
ways to test the EMH. Unfortunately, Fama’s early focus on the random walk hypothesis (and the early
empirical success of the random walk hypothesis) has led to confusion about whether or not markets are
efficient. In what follows I will present Fama’s early presentation of the EMH. I will then follow-up with
a discussion of the random walk hypothesis, the lack of support for it in the data, and why the failure of
the random walk hypothesis is not a failure of the EMH.
Let pt denote the price of an asset at time t and rt denote the percentage one period rate of return
at time t. Fama suggests that the EMH implies that

xj,t+1 = pj,t+1 − E(pj,t+1 |Φt ) (72)

E(x̃j,t+1 |Φt ) = 0 (73)

7
This story is described in greater detail with more context in the excellent book by Mehrling (2005).

21
where xt+1 is the excess profit earned given one’s expectation, Φt is the information available at time
t. Given this definition, what the first equation says is that given the information available at time t,
individuals form an expectation of what the price will be at time t + 1. If the price is higher than the
expected price, then the individual will make a profit. If the price is lower than the expected price, then
the individual will take a loss. The second condition outlines the EMH. It says that given the information
available at time t, the expectation is that pj,t+1 = E(pj,t+1 |Φt ). So what determines E(pj,t+1 |Φt )? And
how can we test this theory?
Let’s start with what this condition actually says. What it says is that individuals must use the
information they have available at time t to form an expectation about what the price will be in the
future such that they expect that they get the price correct. So how does one form an expectation? Well,
typically you write down an asset pricing model. Then, given the information available at time t, you
form an expectation about what the price will be at time t + 1 based on your model. If you have the right
model, then your expectation should be correct (on average).
This insight alone, however, makes it impossible to test the EMH directly. To understand why,
consider the following thought experiment. Suppose that you write down an asset pricing model that you
think fits the data. Now, what you want to do is see if your model fits out-of-sample. So you go to the
data and you see how well your model predicts the behavior of asset prices. If your model fits the data,
then you might be able to conclude that the market outcome is consistent with your model of market
equilibrium. This would provide evidence in favor of the EMH. However, suppose that when you take
your predictions to the data you find that your predicted prices are systematically different than the actual
observed prices. Can you reject the efficient markets hypothesis? No. The reason that you cannot reject
the EMH is that you are testing a joint hypothesis. When you test whether your theory explains the data,
you are testing the joint hypothesis that that the equilibrium price from your model is correct and that
the market is efficient. If the data evolve differently than your model of market equilibrium then either
your theory is wrong or the market is inefficient. In other words, rejecting a test of the EMH implies
one of 3 possibilities: (1) your model provides the “correct” expectation and the market consistently gets
things wrong (i.e. the market is inefficient), (2), your model is wrong, or (3) the market is inefficient and
your model is wrong. However, we have no way to distinguish between these explanations.
This debate was complicated by the fact that some of the earliest theories of asset prices used to test
the EMH made the assumption that prices were martingales. Formally, this implies that E(pj,t+1 |Φt ) =
pj,t . In other words, if asset price are martingales, this implies that the only relevant information for
predicting next period’s price is this period’s price. Plugging this into equation (72), and given the
assumption that E(xj,t+1 |Φt ) = 0, implies that stock prices follow a random walk. This idea dates back
to Bachelier (1900), who reasoned that market speculation must be a fair bet and therefore stock prices
would follow a Brownian motion. Others, like Samuelson (1965), also argued that stock prices would
follow a random walk if all available information was incorporated as well.8
8
Mandelbrot (1963) argued stock prices follow a Paretian distribution. This was motivated by the fact that Mandelbrot had
observed that the distribution of changes in stock prices had “too high of a peak” to be drawn from a normal distribution.
He argued that a Paretian distribution did a much better job of fitting the data. Fama (1965), a student of Mandelbrot at
Chicago, suggests that stock prices are still martingales in this context in the sense that changes in prices can be thought of
as independent draws from a Paretian distribution rather than a normal distribution (as the assumption of Brownian motion
would imply). However, Mandelbrot (1966) wrote a follow-up paper in which he argued that the process by which a Paretian
distribution emerged could imply that changes in stock prices were not martingales. Whether or not stock prices are martingales
when changes in stock prices are not drawn from a normal distribution depends on the process by which we get to equilibrium.
For example, Thompson and Hickson (2006) suggest that we can think of prices as being given by the following learning
process, pt = pt−1 + f [x(pt )], where x is the excess demand of the asset as a function of price. This suggests that the price
this period is equal to the price last period plus some learning function, which is a function of the excess demand for the asset
and therefore the asset’s price. Assume that f (0) = 0 and f 0 > 0. Thompson and Hickson propose the following tatonnement
process. Suppose that there is excess demand for the asset at time t. Then the individual will construct a first-order Taylor

22
 Early tests of market efficiency were actually tests of this random walk model. Much of the research
sought to test whether stock prices followed a random walk using both direct and indirect methods. Using
indirect methods meant using data that was available before time t to see whether this information could
predict changes in the price. If so, this would provide some evidence against the random walk hypothesis
since the random walk hypothesis implies that the best guess of next period’s price is the current period’s
price. Other tests tried to test whether or not stock prices had a unit root. Finally, another way to test
whether or not the data followed a random walk was to do a test on the variance. The random walk
assumption implies that the variance is increasing proportionately with time. Lo and MacKinlay (1987)
use these variance tests and found that the variance of stock prices was actually larger than one would
predict under the null hypothesis of a random walk. This finding has turned out to be quite robust.
Because of the use of the random walk hypothesis at Chicago and Fama’s application of the random
walk hypothesis to test the EMH, one must be careful not to mistake a rejection of the random walk
hypothesis for a rejection of the EMH. The random walk hypothesis can be wrong even if markets are
efficient. All the rejection of the random walk hypothesis implies is that this equilibrium concept is
incorrect. For example, suppose that we replace the random walk hypothesis with the Lucas tree model.
In that model, in a competitive equilibrium the price of an asset is equal to the expected discounted
value of future dividends. This gives us a testable hypothesis. Alternatively, Ross’s APT model assumes
that the relevant equilibrium concept is the absence of risk-free arbitrage. This assumption gives us a lot
of testable models. For example, the CAPM is a testable, one-factor model that is a special case of APT.
The rejection of CAPM therefore does not rule out the equilibrium concept used by Ross, but rather the
CAPM special case. Ross’s APT suggests that there is a seemingly infinite number of factor models out
there that could explain prices. In other words, the way to interpret the EMH in the context of Ross’s
APT is to say that the EMH implies that there are no opportunities for risk-free arbitrage.
This is one reason that I like Ross’s model both as a pedagogical device and as a practical model.
As we stated above, when testing the EMH, you are necessarily testing a joint hypothesis. This requires
determining whether actual market prices are consistent with an equilibrium concept. However, from
an empirical point of view, there is no reason to believe that at any moment in time we are actually in
equilibrium. Markets are a discovery process. Each individual investor might have his or her own theory
about what determines asset prices. These individuals then enter the market and try to profit from these
ideas. In short, all of these investors could be thought of as “seeking alpha.” In other words, suppose
that you take a factor model to the data. For simplicity, suppose that factor model is the CAPM. You
could construct various portfolios using criteria that you think might provide you with excess returns.
You could then estimate the following model for your various portfolios:

(ri,t − rt ) = α + β(rm,t − rt ) + et

where ri is the rate of return on portfolio i, rm is the rate of return earned by the market, and r is the
rate of return on the risk-free asset. This model implies that aggregate risk is perfectly capture by the
market factor. This implies that that if your model is correct, α = 0. Thus, if you estimate the model and
find that for some portfolios, α > 0, then these portfolios earn an excess return above the risk-free rate
that is not explained by your risk factor. If you really believe the CAPM is the appropriate model, then
series approximation of f and solve. If the market clears, then x = 0. Regardless, the price jumps in accordance with this
approximation. If there is still excess demand for the asset, then the individual will construct a second-order Taylor series
approximation of f and solve for f using numerical methods. Again, this causes a jump in the price because it is a refinement
of the previous attempt. If x = 0, then we are in equilibrium. If not, the individual will construct a third-order Taylor series
approximation and solve f using numerical methods. The price jumps either to the equilibrium price or the process repeats.
What this implies is that the prices “jump” with each iteration rather than smoothly adjusting to get to equilibrium. In this
case, asset prices will be convergent martingales, but the error term will exhibit jumps that are inconsistent with a normal
distribution.

23
this result implies there are opportunities for risk-free arbitrage. For example, you could short a portfolio
with α = 0 and use the proceeds to buy a portfolio with the same β, but α > 0. Assuming your model
is correct, you’ve discovered a risk-free money-making machine.
Nonetheless, let’s take a step back and consider the implications of your discovery. Suppose that
you identify these portfolios in which α > 0. There are 4 possibilities here: (1) you have identified a
risk-free arbitrage opportunity, but this opportunity will be short-lived as the availability of this oppor-
tunity becomes more widely known and α is driven down to 0; (2) your results are spurious; (3) your
discovery that α > 0 will provide persistent excess returns, but this is because you are unknowingly
bearing additional risk (i.e., what you have identified is not risk-free arbitrage, but rather is a source of
returns that is explained by some other risk factor that you have left out of your model; your model is
wrong/incomplete); or (4) the market is inefficient and you can continue to earn these excess returns.
Markets are a discovery process in which we can put our model to the test. However, suppose that
we take seriously the notion of the absence of risk-free arbitrage. This suggests that if (1) is true, you will
be able to profit from risk-free arbitrage, but others will discover this opportunity. As more and more
people discover the opportunity, prices will not deviate as far from the price implied by your model as
everyone tries to be the first to capitalize on this opportunity and therefore the opportunity for profit
disappears. If you see that α is being driven to 0, this is evidence that you are in scenario (1). In this
case, using Ross’s methodology, you were able to identify a risk-free arbitrage opportunity, but just as
his model predicts, this opportunity disappears as we move toward equilibrium. This does not violate
market efficiency.
If scenario (2) is correct, the arbitrage opportunity is spurious. This means that you quickly realize
that α ≤ 0 because you either aren’t earning the expected returns you anticipated or you lose enough
money with this false belief that you have to leave the market. This does not violate market efficiency.
If α is not driven to 0, then this implies that either scenario (3) or scenario (4) is correct. In other
words, you are either unknowingly bearing additional risk and you do not realize it, or the market is
inefficient and you have identified a portfolio (or portfolios) that allow you to capitalize on this inefficiency.
There are essentially two ways of determining which scenario that you are in. The first is to search for
additional ways to measure risk. This means trying to identify additional factors that could explain these
excess returns. If you amend the model to include additional factors and find that α = 0 once you
include the additional (or different) factors, then this suggests you are in scenario (3). Nonetheless, this is
not a great test since the absence of evidence is not the evidence of absence. In other words, the failure
to identify additional factors that can drive α to 0 does not imply that such factors do not exist.
So how can we figure out if we are in scenario (3) or (4) since they each have different implications
for market efficiency? One way to do so is to recognize that markets are a discovery process. If we “let
a thousand models bloom” and investors are all interacting in markets and attempting to exploit what
they believe to be risk-free arbitrage opportunities, then the profit and loss mechanism should eliminate
any risk-free arbitrage opportunities. One indirect way to determine that the market is efficient in this
context would be to compare the performance of stock pickers, money managers, etc. to the performance
of the overall market. If a sufficiently large fraction of individuals can outperform a broad market index
such that the number of outperformers is larger than could be predicted by sheer randomness, then
this would be evidence against market efficiency. Even this is an imperfect test because it will tend to
overstate the degree to which investors can beat the market in a risk-adjusted sense since some of the
strategies might consistently and persistently beat the market because they require taking on additional
risk in comparison to the market.9 Nonetheless, some such as Burton Malkiel (2003) have used this as
their preferred test of market efficiency. Malkiel tracks 355 mutual funds from 1970 - 2001. He finds that
of the 355 funds that started the period, 197 of the funds did not survive the entire sample and of the
9
Furthermore, such a comparison might overstate money manage performance if one only looks at the surviving firms.

24
remaining 158 funds, 86 of the funds underperformed the market, and 50 provided market equivalent
returns, and 22 outperformed the market. Nonetheless, even among the 22 firms that outperformed the
market, only 5 of these firms outperformed the market by 2 percent per year or more. Malkiel (2003:
p. 78) concludes that “managed funds are regularly outperformed by broad index funds, with equivalent
risk.” This seems to provide fairly strong evidence in favor of market efficiency.10

7.2 Some Misconceptions

Despite our discussion, debates surrounding the EMH remain the subject of some confusion. For example,
consider the following quote from Shleifer (2000, p. 10):11
Investors follow the advice of financial gurus, fail to diversify, actively trade stocks and churn
their portfolios, sell winning stocks and hold on to losing stocks thereby increasing their tax
liabilities, buy and sell actively and expensively managed mutual funds, follow stock price
patterns and other popular models. In short, investors hardly pursue the passive strategies
expected of uninformed market participants by the efficient markets theory.
Let’s think about this in terms of the efficient markets hypothesis. What the EMH says is that

xt+1 = pt+1 − E(pj,t+1 |Φt )

E(x̃t+1 |Φt ) = 0
where E(pj,t+1 |Φt ) is given to us by a model and x̃ is the difference between the actual price and the
expected price, given that some model has been used to form this expectation. The EMH makes no
assumption about what model you choose to generate this expectation. Thus, if Shleifer’s critique is
about the silly things that investors do and that these silly things are kept out of the models used to
test the EMH, then his critique is of the models and not the EMH. The EMH itself is not a theory, it
is a hypothesis. However, what Shleifer is often referring to in his book when he critiques the idea of
efficient markets are the models put forth by Fama in his 1970 paper to test the EMH, not the EMH itself.
Many of his critiques, for example, represent critiques of the “weak-form” hypothesis put forth by Fama
that stocks follow a random walk. However, even Fama has admitted that random walk models do not
describe the data as well as they did when he wrote his paper. Furthermore, as Fama points out in his
Nobel Prize lecture, the assumption that E(pj,t+1 |Φt ) = pt is just one possible model that can be used
to test the EMH. The failure of the random walk model is not a failure of EMH.
Furthermore, there seems to be some confusion about the distinction between efficiency as a general
concept and the EMH. One way to think about efficiency is as follows. Suppose that there is some true
rate of return, rT . We might define an efficient market as one in which the actual rate of return, r, is
equal to the true rate of return rT . If we assume that information is costless, then it is almost trivial to
argue that r = rT in equilibrium. A number of advocates of efficient markets argue that the ability to
arbitrage will be sufficient to get r = rT . The idea that arbitrage will get us to the true price is significant
because the mere presence of arbitrage would seem to rule out the importance of any of the silly behavior
of investors described in the previous Shleifer quote. In other words, suppose that the people doing silly
things are systematically silly (i.e., they are not only silly, but they are silly in a particular way such that
10
Malkiel is effectively replicating a similar analysis as Jensen (1968).
11
In what follows, I am going to use Shleifer as my point of comparison. He is not alone in the wilderness with respect to
his point of view. In addition, despite using Shleifer to compare and contrast notions of efficiency, this discussion should not be
seen as a criticism of Shleifer. In fact, if I wasn’t interested in his work on noise trading, I never would have wrestled with these
arguments in the first place! Nonetheless, I use Shleifer because I think that he provides the clearest description of the view
that markets are “inefficient.” I also think this clarity helped me to realize that when he says “efficiency,” he means something
different than Fama.

25
their behavior does not “cancel out.”). All it would take is for one or two non-silly people to enter the
market. These people could use arbitrage to profit from the silliness of everyone else. Eventually, this
arbitrage process would result in r = rT . Shleifer (2000, p. 13) presents a critique of this view:

This brings us to the ultimate set of theoretical arguments for efficient markets, those based
on arbitrage. Even if sentiment is correlated across unsophisticated investors, the arbitrageurs
– who perhaps are not subject to psychological biases – should take the other side of unso-
phisticated demand and bring prices back to fundamental values. Ultimately, the theoretical
case for efficient markets depends on the effectiveness of such arbitrage.
The central argument of behavioral finance states that, in contrast to the efficient markets
theory, real-world arbitrage is risky and therefore limited.

Note, however, that his critique is that arbitrage will not get us to a point in which price is equal to
fundamental value because there are risks associated with arbitrage and therefore there are limits to
arbitrage. The idea that price might not equal fundamental value seems like a rather strict definition of
efficiency, especially given the emphasis on subjective valuation in economics. Nonetheless, even if this
is one’s view of efficiency, it is not the argument made by the EMH. The EMH is actually a much weaker
definition of efficiency. What the EMH says is that we should expect that, on average,

E(rj,t+1 |Φt ) = rj,t+1

where rj,t+1 is the actual rate of return at time t + 1 and E(rj,t+1 |Φt ) is the expected rate of return at
time t + 1 given the information available at time t. This expected rate of return comes from a model.
This could be the CAPM, the C-CAPM, our arbitrage model, or any other model. I will focus on the
arbitrage model because of Shleifer’s own emphasis on arbitrage.
In our arbitrage pricing model, we argued that there should be an absence of risk-free arbitrage. In
other words, suppose that we believe that the CAPM version of our arbitrage pricing model is correct.
This implies that the expected returns on a portfolio should be given as

µp = r + β(µM − r)

where µp is the expected return on the portfolio, µM is the expected return on the market, r is a risk-free
return, and β measures the sensitivity of the expected returns of the portfolio to the market. Now suppose
that I construct another portfolio. Let µ̃p be the expected returns on this second portfolio and suppose
that
µ̃p = r + γ + β(µM − r)
Note that the risk factor is identical. Thus, if γ > 0, I would earn a higher expected return on the second
portfolio despite the same amount of risk. I could therefore produce a self-financing portfolio in which I
short the first portfolio, use the proceeds from the sale to buy the second portfolio and earn an expected
rate of return, µ̃p − µp = γ. This portfolio is self-financing in the sense that I don’t have to use any of my
own funds and I can earn a positive expected rate of return. Since each portfolio has the same degree
of aggregate risk and no idiosyncratic risk, I can completely eliminate risk by buying one portfolio and
selling the other. It follows that I can earn γ risk-free. If so, then everyone would have an incentive to
capitalize on this risk-free money-making opportunity. In doing so, purchasing the second portfolio and
selling the first portfolio should reduce the relative rate of return on the second portfolio until γ = 0.
In the context of the CAPM this implies that portfolios that have the same β (i.e. the same sensitivity
to market risk), should have the same expected rate of return. Otherwise, investors could earn risk-free
arbitrage profits. Even if we extend this model to multi-factor models, the implication is the same. For

26
example, instead of CAPM, we could think of another factor model consistent with the APT, such as
Chen-Roll-Ross, where the expected rate of return on one portfolio is given as
µp = r + βp,IP (µIP − r) + βp,U I (µU I − r) + βp,EI (µEI − r) + βp,T S (µT S − r) + βp,RP (µRP − r)
If there is another portfolio with the same risk characteristics with an expected rate of return
µ̃p = r + γ + βp,IP (µIP − r) + βp,U I (µU I − r) + βp,EI (µEI − r) + βp,T S (µT S − r) + βp,RP (µRP − r)
then it follows that γ must be equal to zero here as well. Otherwise, there would be an opportunity for
risk-free arbitrage profits.
The moral of the story with regards to our arbitrage pricing model is not that there will be an
absence of all arbitrage opportunities. Rather, our model implies that there will be an absence of risk-
free arbitrage opportunities. In other words, if there are differences in expected rates of return that do
not reflect differences in risk premia, then arbitrageurs will step in and earn risk-free arbitrage profits.
As a result, in equilibrium, differences in expected rates of return should be entirely explained by risk
premia.
Consider a simple thought experiment. Suppose that you observe a stock that trades on two different
exchanges and the price is different on the exchanges. This is an arbitrage opportunity because one
could profit from buying the stock cheaply on one exchange and selling the stock short on the expensive
exchange. However, this is not necessarily risk-free arbitrage. In fact, Shleifer admits as much, arguing
that the lack of arbitrage that gives rise to these deviations is due to noise trader risk.
So why does Shleifer say that markets are inefficient? Shleifer is making the case that markets are
inefficient because there are various frictions in the market that prevent the price from getting to its
“frictionless price,” or “fundamental value.” But this creates confusion because on the one hand we have
a theoretical notion of efficiency and we have an empirical hypothesis, the EMH, that are using the same
word, “efficient,” to mean two different things. Shleifer is saying that markets are inefficient because there
are various frictions like incomplete markets, imperfect information, and noise traders make it so that
the price will not equal its fundamental value. However, the EMH just says that the market is efficient if
the expected price is, on average, equal to the actual price.
In fairness to Shleifer, he is not likely to agree with my characterization. For example, in his work he
differentiates between what he calls fundamental risk and the risk associated with investor sentiment. To
understand this position, we need to discuss the concept of noise.

7.3 Noise
According to Fischer Black (1986), noise is what “makes financial markets possible, but also [what] makes
them imperfect.” What Black was referring to was the fact that people might sell their assets for reasons
that have nothing to do with information that would be relevant for valuation. For example, people
might sell stocks because they need money to make a purchase. In addition, when buying an asset, there
is one buyer and one seller. Abstracting from liquidity needs, if they share the same information and
information was all that mattered, then one of these individuals would be making a mistake. However, if
one person is making a mistake and they know it, then trade would never take place. So the only way
that the trade takes place is if the trade is based on something other than the shared information. Noise
can therefore be thought of as something akin to beliefs. As Black notes, this definition is complicated
by the fact that beliefs are often a function of information. Thus, a more appealing definition of noise
traders suggested by Black is that noise traders often think they are trading on information, but they are
not (i.e. it is hard to parse what is information and what it noise).
So why does any of this matter? Well, if people are selling stocks to generate liquidity or based
on beliefs, this will tend to put noise into the price of assets. Whether this matters depends on the

27
cumulative effect of noise. For example, suppose that in the absence of noise traders there would be
some information-based price (i.e. information traders know the “true” model of asset prices). If noise
traders make cumulative errors, then the noise introduced into the price would cause the price to deviate
quite significantly from the information-based price. These cumulative errors present a profit opportunity
for information traders. If so, wouldn’t information traders profit from the mistakes of noise traders and
push the price back toward the information-based price? Not necessarily. In the real world, it might be
hard to determine who the information traders and the noise traders really are. Some noise traders might
erroneously think they are information traders. As a result, the large positions necessary to get maintain
prices that are free of noise would entail significant risk. Black therefore argued that while noise traders
increase the liquidity of stock markets, there is no guarantee that more active trading leads to greater
efficiency in pricing. Put simply, “noise creates the opportunity to trade profitably, but at the same time
makes it difficult to trade profitably” (Black 1986, p. 534).
So what does this imply about market efficiency and the EMH?
In his paper, Black seems to think a bit more carefully about this than most. He adopts the standard
theoretical definition that the correct price is the frictionless price (i.e. the price in the absence of noise
traders). Using this definition, he states that “what’s needed for a liquid market causes prices to be less
efficient” (Black 1986, p. 532). However, Black was careful to differentiate between theoretical efficiency
and empirical efficiency because “value is not observable” (Black 1986, p. 533). As a result, Black (1986,
p. 533) argues that
. . . we might define an efficient market as one in which price in within a factor of 2 of value,
i.e., the price is more than half of value and less than twice value. The factor of 2 is arbitrary,
of course. Intuitively, though, it seems reasonable to me, in the light of sources of uncertainty
about value and the strength of the forces tending to cause price to return to value. By this
definition, I think almost all markets are efficient almost all the time.
What Black is arguing is that the mere presence of noise traders is what allows us to have highly liquid
financial markets because noise traders will trade more frequently than information traders. In theory, if
we could divide traders into information traders and noise traders, then the information traders would
tend to push prices toward the frictionless price. Noise traders would tend to push prices away from
the frictionless price if their errors are cumulative. Thus, in theory, the presence of noise traders can
create an inefficiency in the sense that they push the price away from the frictionless price. However, in
practice, there are two limitations that are absent from theory: (1) we do not know the fundamental value
of an asset, and (2) we do not know which traders are noise traders and which traders are information
traders. In fact, all noise traders likely believe that they are information traders! What this means is that
while noise traders might prevent prices from converging to the frictionless price in theory, this does not
necessarily violate the EMH in practice.
Black implicitly recognized this. What Fischer Black was arguing with his “factor of 2” idea for
empirical testing is that noise traders introduce irreducible risk into market. Noise traders are a source
of aggregate risk. To understand this, consider a simple thought experiment. Suppose that the CAPM
would perfectly predict stock prices in a world with no noise traders. In a world with noise traders,
prices can deviate significantly those predicted by CAPM. However, suppose that you are using CAPM.
You think that CAPM is the appropriate model, but you don’t know that for certain. You might be one
of the noise traders that everyone is complaining about. So when prices deviate from those predicted
by CAPM, there is risk associated with trying to profit from the deviation because you do not know for
certain that the trade will be profitable. Furthermore, even if you knew with absolute certainty that the
CAPM was the best way to price stocks, there is still the risk that noise traders can continue to push the
price in the wrong direction for an extended period of time. The question of how to test market efficiency
when people have heterogeneous beliefs, some of those beliefs are not tied to relevant information, and

28
we do not know what information is relevant is a difficult on to say the least. Black’s “factor of 2” method
is just one way of handling the empirical problem of testing market efficiency in the presence of noise
trading.12
Shleifer takes a different view. In Shleifer’s view, if noise traders make cumulative errors, they are
pushing the stock price away from its frictionless price and therefore markets are inefficient. However,
when using this idea for empirical work, this presumes that (a) some frictionless price exists, and (b) this
frictionless price can be known. This is a stronger definition of efficiency than the EMH. In contrast,
something like the APT just assumes the absence of risk-free arbitrage (i.e., the rate of return on two
portfolios with the same risk will have the same expected rate of return). This framework doesn’t presume
that we know or could ever know the frictionless price, or “fundamental value” of a stock. A test of the
EMH using an APT model is a test of whether the market is efficient and whether the factor model
constructed using the APT. An efficient market in this context is a market in which there is no risk-free
arbitrage. This is not the same as saying that the actual price is equal to the frictionless price. In fact, the
APT is a much more modest version of efficiency that implies that the best we can do is look for factors
that explain returns. There is no attempt to model the frictionless price. This distinction is important
because Shleifer argues that we can think of risk in a noisy market as being the risk associated with
fundamentals or the risk associated with investor sentiment (noise). In an APT model, risk is risk, no
matter the source, as long as we can create a tracking portfolio. If we can construct a tracking portfolio
of investor sentiment, then we could use the APT to derive expected returns consistent with this risk
factor just like any other risk factor. It should be noted that this is a big “if.” The inability to do so
undermines the APT (assuming sentiment is a legitimate source of risk). Whether or not it undermines
the EMH, however, is ambiguous.
When it comes to noise, it seems that efficiency is in the eye of the beholder.

7.4 The Bottom Line

The following is a basic summary of the EMH:

• The EMH is an empirical hypothesis that on average we should expect E(pj,t+1 |Φt ) = pj,t+1 ,
where pj is the price of asset j and Φt is the information available at time t.

• Since we must have a model of asset prices in order to form an expectation about future asset
prices, the EMH is a joint hypothesis. Put differently, the EMH is the hypothesis that both (a) your
model is the correct model, and (b) the market is efficient.

• The EMH does not require that stock prices follow a random walk.

• How we should interpret what the EMH means depends on our theory of asset prices. Under the
random walk theory, an efficient market is one in which stock prices are unpredictable. Under
CAPM, an efficient market is one in which expected excess returns should be explained solely by
expected excess returns of the market and the co-movement of the asset price with the market.
Under the C-CAPM, an efficient market is one in which the expected excess returns of an asset
are entirely explained by the covariance of consumption growth and the asset’s return. Under the
APT, an efficient market is one in which there are not opportunities for risk-free arbitrage. In
other words, how narrow or how broad we define efficiency depends on our choice of asset pricing
model. The APT is quite flexible; so flexible that indirect tests, such as those of Jensen and Malkiel,
are likely more appropriate than any test of a specific variant of APT as it relates to the EMH.

12
Mehrling (2005) argues that Black’s use of noise and noise traders was to reconcile his affinity for the CAPM with the data.

29
Appendix: A Binomial Representation of Brownian Motion
We will actually deal with geometric Brownian motion because we can see how this would apply to
geometric Brownian motion and arithmetic Brownian motion in one result.
Suppose that x follows a geometric Brownian motion

dx
= µdt + σdz (74)
x
Now suppose that y = f (x) = ln x. Using Ito’s Lemma, we have
1
dy = f 0 (x)dx + f 00 (x)(dx)2
2
1 1 1
dy = dx − 2 (dx)2 = µdt + σdz − 2 σ 2 dt
x x x
Or,
dy = αdt + σdz
where  
1 2
α= µ− σ
2
Now, let’s show how this expression has a binomial representation.
Consider a discrete change in y during one unit of time, which will we denote ∆y. Suppose that ∆y
is drawn from a binomial distribution such that
w/probability p

γ
∆y =
−γ w/probability 1 − p

It follows that
E(∆y) = pγ − (1 − p)γ = [p − (1 − p)]γ = (2p − 1)γ
Furthermore, from the definition of variance, we have

var(∆y) = E[(∆y)2 ] − E[∆y]2

So,

var(∆y) = pγ 2 + (1 − p)(−γ)2 − (2p − 1)2 γ 2 = γ 2 − (4p2 − 4p + 1)γ 2 = 4p(1 − p)γ 2

To approximate the Brownian motion, we need the following to be true:

(2p − 1)γ = αdt

4p(1 − p)γ 2 = σ 2 dt
To simplify this, let’s square both sides of the first expression to get

(2p − 1)2 γ 2 + 4p(1 − p)γ 2 = α2 (dt)2 + σ 2 dt

As is our convention, we will ignore terms of higher order than dt such that

(2p − 1)2 γ 2 + 4p(1 − p)γ 2 ≈ σ 2 dt

30
Some basic algebraic manipulation implies that this is equivalent to

γ 2 = σ 2 dt

So √
γ = σ dt
Plugging this back into the first condition yields

(2p − 1)σdt1/2 = αdt

Solving for p,
α√
 
1
p= 1+ dt
2 σ
Now, let’s return to our discrete approximation. One way to think about this discrete approximation is to
say that
yt−1 + γ w/probability p

yt =
yt−1 − γ w/probability 1 − p
However, recall that y = ln x, which implies that x = ey . We can therefore write:

w/probability p
 γ
e xt−1
xt = (75)
e xt−1 w/probability 1 − p
−γ

So, we can now show how to relate this to our discrete time option model above. Suppose we have a
stock price St that follows a geometric Brownian motion:

dS
= µdt + σdz (76)
S
It follows that we can write:
uSt−1 w/probability p

St = (77)
dSt−1 w/probability 1 − p
where √
u = eσ dt
√
d = e−σ dt

µ − (σ 2 /2) √
 
1
p= 1+ dt
2 σ

31
References
[1] Bachelier, Louis. 1900. Theorie de la Speculation. Doctoral Dissertation, University of Paris.

[2] Bjork, Tomas. 2009. Arbitrage Theory in Continuous Time. Oxford: Oxford University Press.

[3] Black, Fischer. 1986. “Noise.” Journal of Finance, Vol. 41, No. 3, p. 529 - 543.

[4] Black, Fischer and Myron Scholes. 1973. “The Pricing of Options and Corporate Liabilities.” Journal
of Political Economy, Vol. 81, p. 659 - 683.

[5] Chen, Nai-Fu, Richard Roll, and Stephen A. Ross. 1986. “Economic Forces and the Stock Market.”
The Journal of Business, Vol. 59, No. 3, p. 383 - 403.

[6] Cox, John C., Stephen A. Ross, and Mark Rubinstein. 1979. “Option Pricing: A Simplified Ap-
proach.” Journal of Financial Economics, Vol. 7, p. 229 - 263.

[7] Derman, Emanuel. 2002. “The Perception of Time, Risk, and Return During Periods of Speculation.”
Quantitative Finance, Vol. 2, p. 282 - 296.

[8] Derman, Emanuel and Nassim Nicholas Taleb. 2005. “The Illusions of Dynamic Replication.” Quan-
titative Finance, Vol. 5, No. 4, p. 323 - 326.

[9] Fama, Eugene F. 1965. “The Behavior of Stock Market Prices.” Journal of Business, Vol. 38, p. 34 -
105.

[10] Fama, Eugene F. 1970. “Efficient Capital Markets: A Review of Theory and Empirical Work.” Journal
of Finance, Vol. 25, No. 2, p. 383 - 417.

[11] Fama, Eugene F. 2013. “Two Pillars of Asset Pricing.” Nobel Prize Lecture, December 8, 2013.

[12] Gabaix, Xavier. 2012. “Variable Rare Disasters: An Exactly Solved Framework for Ten Puzzles in
Macro-Finance.” Quarterly Journal of Economics, Vol. 127, p. 645 - 700.

[13] Gardner, John and Joshua R. Hendrickson. 2017. “If I Leave Here Tomorrow: An Option View of
Migration When Labor Market Quality Declines.” Southern Economic Journal, Forthcoming.

[14] Hendrickson, Joshua R. and Alexander William Salter. 2016. “A Theory of Why the Ruthless Revolt.”
Economics & Politics, Vol. 28, No. 3, p. 295 - 316.

[15] Jensen, Michael C. 1968. “The Performance of Mutual Funds in the Period 1945 - 1964.” The Journal
of Finance, Vol. 23, No. 2, p. 389 - 416.

[16] Lo, Andrew W. and A. Craig MacKinlay. 1988. “Stock Prices Do Not Follow Random Walks.” The
Review of Financial Studies, Vol. 1, No. 1, p. 41 - 66.

[17] Malkiel, Burton G. 2003. “The Efficient Markets Hypothesis.” Journal of Economic Perspectives, Vol.
17, No. 1, p. 59 - 82.

[18] Mandelbrot, Benoit. 1963. “The Variation of Certain Speculative Prices.” Journal of Business, Vol. 36,
p. 392 - 417.

[19] Mandelbrot, Benoit. 1966. “Forecasts of Future Prices, Unbiased Markets, and Martingale Models.”
Journal of Business, Vol. 39, p. 242 - 255.

32
[20] Mehra, Rajnish and Edward C. Prescott. 2003. “The equity premium in retrospect.” Handbook of the
Economics of Finance.

[21] Mehrling, Perry. 2005. Fischer Black and the Revolutionary Idea of Finance. New Jersey: Wiley & Sons.

[22] Merton, Robert C. 1973. “The Theory of Rational Option Pricing.” Bell Journal of Economics and
Management Science, Vol. 4, p. 141 - 183.

[23] Ross, Stephen A. 1976. “The Arbitrage Theory of Capital Asset Pricing.” Journal of Economic Theory,
Vol. 13, p. 341 - 360.

[24] Samuelson, Paul A. 1965. “Proof That Properly Anticipated Prices Fluctuate Randomly.” Industrial
Management Review, Vol. 6, p. 41 - 49.

[25] Shleifer, Andrei. 2000. Inefficient Markets: An Introduction to Behavioral Finance. Oxford: Oxford
University Press.

[26] Thompson, Earl A. and Charles R. Hickson. 2006. “Predicting Bubbles.” Global Business and Eco-
nomics Review, Vol. 8, No. 3/4, p. 217 - 246.

33
 Investment Notes
Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
An understanding of the causes and consequences of capital accumulation is important for several
reasons. First, the study of economic growth often involves a role for capital accumulation. The extent
to which capital accumulation influences economic well-being is therefore of some importance. Second,
there are significant business cycle properties of investment. Recall from Principles of Macroeconomics
the National Income Identity:
Y =C +I +G+X
where Y is GDP, C is consumption, I is investment, G is government purchases, and X is net exports. If
we look at each component, investment (i) has the highest contemporaneous correlation with real GDP,
(ii) is the most volatile component, and (iii) is much more volatile than real GDP (Cooley and Prescott,
1995).
A time-series graph of real private, non-residential fixed investment is shown in Figure 1.

Real private fixed investment: Nonresidential (chain-type quantity

index)

140

120

100
(Index 2009=100)

80

60

40

20

0
1950 1960 1970 1980 1990 2000 2010

Source: US. Bureau of Economic Analysis

Shaded areas indicate US recessions - 2015 research.stlouisfed.org

Figure 1: Real Private Non-Residential Fixed Investment

Investment itself consists of business fixed investment, residential investment, and inventory invest-
ment. For our purposes, we will focus much of our time on business fixed investment. In particular,

1
we are interested in the following questions. How do firms make investment decisions? What factors
are important? What factors are unimportant? How do fixed costs affect investment decisions? Are
borrowing constraints important?

2 The Accelerator Model

One of the earliest models of investment is the “accelerator model” of Clark (1917), who inverted a
production function, took first differences, and arrived at a specification for investment. Not surprisingly
given its simplicity, this theory had a hard time capturing important characteristics of the data. For
example, as shown in Figure 2, there is strong serial correlation of investment. The accelerator model
could not explain the persistence of investment.
1.00
Autocorrelations of investment
-0.50 0.00
-1.00 0.50

0 5 10 15 20 25
Lag
Bartlett's formula for MA(q) 95% confidence bands

Figure 2: Autocorrelation of Real Private Non-Residential Fixed Investment, 1947:Q1 - 2015:Q1.

The inability to explain serial correlation in investment led to the subsequent development of the
flexible accelerator model (Clark, 1944; Koyck, 1954). This model suggested that investment was given by
N
X
∗
It = ρn ∆Kt−n (1)
n=0
where K ∗ was defined as the desired level of capital and ρn are parameters. Thus, the fundamental idea
is that investment responded to the firm’s desired level of the capital stock and did so gradually.
While this early work was fairly simple, it lacked significant characteristics common to most economic
models. First, the desired level of capital, K ∗ , was assumed to be some constant fraction of output, an
ad hoc assumption. Second, there is an absence of any sort of prices or marginal analysis. These
shortcomings led to the development of Jorgenson’s neoclassical theory of investment and the concept of
the user cost of capital.

3 The User Cost of Capital and the Neoclassical Theory of Investment

In the early 1960s, Dale Jorgenson began to push for a return to the neoclassical theory of investment
that began with Irving Fisher. In Fisher’s approach, the investment decision of the firm was to choose

2
the amount of investment that would maximize the net worth of the firm. Yet, in the 1960s, as Jorgenson
(1963: 247) pointed out “The central feature of the neoclassical theory is the response of the demand for
capital to changes in relative factor prices or the ratio of factor prices to the price of output. This feature
is entirely absent from the econometric literature on investment.” What follows is a description of the
neoclassical theory of investment, as revived by Jorgenson.

3.1 Jorgenson’s Model

Define the net worth of the firm as
Z ∞
W = e−rt [R(t) − D(t)]dt (2)
0

where R is net revenue, D is direct taxes, and r is the real interest rate. The net revenue of the firm is
given as
R = pQ(L, K) − wL − qI (3)
where p is the price of output, w is the wage, q is the price of capital goods, L is the quantity of labor, I
is investment, and Q is the quantity of output, which is assumed to be a function of the quantity of labor
and the capital stock.
Further, define direct taxes as

D = τ [pQ(L, K) − wL − (vδq + srq − xq̇)K]

where τ ∈ (0, 1) is the tax rate, v is the proportion of replacement that is chargeable for income taxes,
s is the proportion of interest chargeable against income, x is the proportion of capital losses that are
chargeable against income taxes, and δ is the constant rate of depreciation.
Substituting the expressions for R and D into the definition of net worth yields a maximization
problem for the firm
Z ∞  
−rt
max e pQ(L, K) − wL − qI − τ [pQ(L, K) − wL − (δq + rq − q̇)K] dt (4)
L,K 0

subject to K̇ = I − δK.
Let’s simplify this problem. Let’s assume that labor is fixed, L = 1. Define Π(K) = (1−τ )pQ(1, K)−
w. We can now write the Hamiltonian for the firm as

H(I, K, λ) = Π(K) − qI − τ [pQ(L, K) − wL − (vδq + srq − xq̇)K] + λ(I − δK)

The conditions that satisfy the Maximum Principle are given as

λ=q

rλ − λ̇ = (1 − τ )pQK + τ (vδq + srq − xq̇) − λδ

K̇ = I − δK
From the first condition, we also know that λ̇ = q̇. Thus, the second condition can be re-written as

rq + δq − q̇ = (1 − τ )pQK + τ (vδq + srq − xq̇)

or  
1 − sτ 1 − τv 1 − τ x q̇ q
QK = r+ δ+ (5)
1−τ 1−τ 1−τ q p

3
Jorgenson defined the term on the right-hand side as the real user cost of capital. Note that the user cost
of capital is increasing in the rate of interest, depreciation, and the tax rate and decreasing in the growth
rate of the price of capital.
Now suppose that s = v = x = 0, this simplifies to
 
1 q̇ q
QK = r+δ− ≡u
1−τ q p

This is the more standard textbook definition of the user cost, which we will denote as u.
Suppose that the firm has a Cobb-Douglas production function, Q = K α L1−α . It follows that the
demand for capital is given as
Q
K=α (6)
u
where u is the real user cost defined above.

3.2 A Neoclassical Flexible Accelerator Model?

Like the original accelerator model, the neoclassical model lacked the ability to replicate the serial cor-
relation of investment data. Thus, Hall and Jorgenson (1967) proposed a neoclassical flexible accelerator
model. Using equation (6) to define K ∗ to be consistent with equation (1), we have flexible accelerator
model in which the desired level of capital is determined by neoclassical theory.

3.3 Limitations
The limitations of this framework are as follows. First, the flexible accelerator modification helps the
model to fit the data, but there is little theoretical motivation for the flexible accelerator. It is basically
just an ad hoc way to fit the model to the data. Second, the model implies that the capital stock should
adjust instantaneously to changes in the user cost. In reality, there are costs associated with capital
accumulation. These costs might be variable (i.e. change with the quantity of investment) or they might
be fixed. We will consider each of these costs below.

4 Tobin’s q
In the 1960s, James Tobin proposed an alternative way of explaining investment behavior. Tobin (1969)
proposed what has come to be known as the q-theory of investment. According to Tobin’s argument,
investment is determined by
I = αq
where α > 0 is a parameter. Here, the q stands for the ratio of the value of the firm relative to the cost
of purchasing the firm’s capital in the current market. The basic idea behind Tobin’s formulation was
that investment should increase as the value of the firm rises relative to the cost of replacing its existing
capital.
Hayashi (1982) made an important contribution to this approach by demonstrating that the addition
of convex adjustment costs to the standard neoclassical model could produce the implications suggested
by Tobin’s q-theory. What follows is Hayashi’s approach.
Let’s consider a problem in continuous time. A firm owns capital, k(t), and uses this capital to
produce its final output good. The price of the capital stock is assumed to be 1. The capital stock of the
firm evolves according to:
k̇ = I

4
where I is investment and we are assuming that capital does not depreciate.
Suppose that the firm’s profit is some function of the aggregate capital stock and also proportional
to the firm’s own capital stock:
Profit = π(K)k
where K is the aggregate capital stock, k is the firm-level capital stock, and π 0 < 0.1
It follows that the firm can increase its profit by increasing its capital stock. However, assume that
the firm faces a convex adjustment cost associated with investment. Denote this cost as C(I), where
C 0 , C 00 > 0.
The firm’s profit maximization problem is given as
Z ∞  
−rt
max e π[K(t)]k(t) − I(t) − C[I(t)] dt
I 0

subject to
k̇ = I(t)
We can write the current-value Hamiltonian as

H(I, k, q) = π(K)k − I − C(I) + qI (7)

where I is the control variable, k is the state variable and q is the co-state variable. The conditions to
satisfy the maximum principle are:
q = 1 + C 0 (I) (8)
rq − q̇ = π(K) (9)
k̇ = I (10)
There are a couple of important points to note from these equilibrium conditions. First, note that in
the steady state, q satisfies
π(K)
q=
r
Thus, q is the marginal value of the firm from increasing its capital stock relative to the marginal cost.
This is Tobin’s marginal q.
Second, combining the first and third conditions above yields

k̇ = C 0−1 (q − 1) (11)

Thus, when q > 1, the firm’s capital stock is growing. When q < 1, the firm’s capital stock is declining.
From our definition of q above, this implies that whenever the marginal value of additional capital exceeds
the marginal cost, the firm will invest. Recall that this was the basic insight behind Tobin’s q.
1
This final assumption is appropriate if we assume that the product produced by the firm has a demand curve that is
downward-sloping. The assumption that the profit are proportional to the firm’s capital stock implies that the production
function has constant returns to scale, output markets are competitive, and the supply of all factors other than capital are
perfectly elastic.

5
5 Lumpy Investment
One characteristic of investment is that it tends to be lumpy (Doms and Dunne, 1998). For example,
during a given period of time, many firms are not investing at all. Much of the investment that takes
place during any given time period is concentrated among a small group of firms. However, our frame-
works thus far have suggested that capital accumulation is gradual and therefore there’s a steady flow
of investment. How can we reconcile this? What are we missing? In this section, we will make one
modification to our framework to generate lumpy investment at the firm level.
Recall that in the framework in the previous section, we assumed that the cost of investment was
I + C(I). Note that this implies that the cost associated with investment and disinvestment is symmetric.
However, this might not be the case. For example, suppose that one must pay a higher price, Pk , for
new capital than they can receive for selling existing capital, pk . It follows that investment is no longer
perfectly reversible.
Given this assumption, we can now define the cost of investment as γ(I) where

 C(I) + Pk I if I > 0


γ(I) = 0 if I = 0
C(I) + pk I if I < 0


where C 0 (I) > 0.

The value of a firm with a given capital stock, K, is therefore given as
1
V (K, t) = max[Π(K) − γ(I)]∆t + V (K 0 , t + ∆t|K)
I 1 + r∆t
where r is the real interest rate, ∆t is an increment of time, and ∆K is a change in the capital stock. The
first term on the right-hand side measures the revenue of the firm as a function of K, where Π0 (K) > 0.
The second term measures the cost of investment. The third term is the present value of the firm from
time t to time t + ∆t.
Multiplying both sides of this expression by (1 + r∆t) and re-arranging yields

r∆tV (K, t) = (1 + r∆t)[Π(K) − γ(I)]∆t + [V (K 0 , t + ∆t|K) − V (K, t)]

Dividing both sides by ∆t and and taking the limit as t → 0 yields a continuous time Bellman equation:
1
rV (K) = max Π(K) − γ(I) + dV
I dt
where dV := lim∆t→0 [V (K 0 , t + ∆t|K) − V (K, t)].
Finally note that the capital stock evolves according to the law of motion:

dK = [I − δK]dt

We can now characterize the optimal behavior of the firm. First, note that since dV = V 0 (K)dK, the
Bellman equation can be written as

rV (K) = max Π(K) + VK [I − δK] − γ(I)

I

It follows that the optimal choice of investment is given satisfies

γ 0 (I ∗ ) = VK

6
Intuitively, this condition implies that the firm should choose a level of investment such that the marginal
value of the additional capital is equal to the marginal cost of investment. However, this condition is
non-linear. For example, note that if investment is positive, the condition is given as

VK = Pk + C 0 (I ∗ )

Thus, the firm should invest if the marginal value of capital is at least as much as the marginal cost. If
VK < Pk + C 0 (I), ∀I > 0, then no investment should take place.
For disinvestment, the condition is

VK = pk + C 0 (I ∗ )

Note that since this is disinvestment, C 0 (I) < 0. This implies that the minimum level of investment that
a firm would be willing to do is I < 0 such that pk + C 0 (I) = 0.
Jointly, this conditions imply the following:

I ∗ ∈ (I, 0) if 0 < VK < pk

I∗ = 0 if pk ≤ VK ≤ Pk
I>0 if VK > Pk

It follows that investment will be lumpy since there is some range over which the marginal value of
capital is not high enough to warrant investment and not low enough to warrant disinvestment. This
is the inaction region. Only when the marginal value of capital leaves this region do firms invest (or
disinvest).

6 The Real Option Theory of Investment

The so-called “real option” theory of investment was developed by a number of researchers utilizing
concepts from financial economics. For a textbook treatment, see Dixit and Pindyck (1994). See also the
work of Dixit (1989) on the entry and exit decisions of firms.2
The basic idea behind the option theory of investment is as follows. Suppose that a firm has the
opportunity to invest in a new project. Denote the profit opportunity of the project as x and assume
that the value is fluctuating over time. In addition, assume that the project is irreversible and that there
is a fixed cost associated with adopting the project. This basic idea bears some resemblance to our
discussion of financial options. For example, suppose that a stock price, X, is fluctuating over time,
but that one can buy a contract that gives one the option to buy the stock at a price, S. This scenario
is similar to the example we just presented in the sense that stock price is analogous to the fluctuating
profit opportunity of the project and the strike price is analogous to the fixed cost of investment. The
one significant difference is that financial options have an expiration date. Often times, the option to
engage in real investment does not have an expiration date (e.g., you don’t have a particular calendar
date before which you have to decide to build a factory). Thus, when discussing real investment, the firm
must decide two things: (1) whether they should exercise the option, and (2) when they should exercise
the option.
Let’s give a concrete example. Suppose that there is a firm that is thinking about building a factory.
Building the factory would allow the firm to produce a particular good (or more of a particular good).
However, the profitability of the project might be fluctuating over time due to things like changes in
demand. Since building the factory is irreversible and requires a large fixed cost, we can think of the
2
Some of the important concepts emphasized in the option theory actually appear in an earlier paper by Bernanke (1983).

7
firm’s investment problem as determining both whether to build the factory and figuring out the optimal
point in time to build the factory. One way to think about this is to think of the firm owning the option
to invest. Like a financial option, the firm has the option, but not the obligation, to exercise the option to
build the factory. Since the investment is irreversible, when the firm exercises the option to invest, it loses
the value of the option. In choosing the optimal point in time to invest, the firm is comparing the value
of the project if they build now with the value of retaining the option to exercise the option to invest
in the future. Put simply, the firm faces a trade-off. The longer the firm waits, the more information it
receives. If this information is favorable, then this increases the value of the project.
We begin by discussing an example of this approach when the value of the project is deterministic.
We then expand to thinking about the problem when the value of the project is stochastic. We then use
a markup approach to discuss the microeconomic interpretation of the investment problem. Finally, we
discuss some extensions.

6.1 A Deterministic Model

Suppose that the profit flow of an investment project is given as x. Suppose that dx = gdt, where g > 0.
Let T denote the time period in which the investment project is undertaken. It follows that for some
initial profit of the project x0 , the profit of the project when the firm invests is egT x0 . A firm that invests
in the project bears a fixed cost I.
Thus, the firm’s investment problem at time t = 0 is given as

max e−ρT [egT x0 − I] = e(g−ρ)T x0 − e−ρT I

T

where ρ > g is the discount rate. Maximizing this expression with respect to T yields

(ρ − g)x0 e(g−ρ)T = ρe−ρT I

The value of T that maximizes this expression is

   
∗ 1 ρI
T = max ln ,0 (12)
g (ρ − g)x0

This expression shows that the optimal choice is choose T to maximize the net present value of the
investment.3

6.2 A Stochastic Control Approach

Consider the same problem that we had in the previous section with the following modification. Instead
of the profit of the project increasing deterministically over time, the profit is evolving stochastically.
The firm’s choice is still the same. However, since the profit of the project is stochastic, the firm can
no longer choose T . Think about why this is the case. What the firm was really doing in the previous
example was waiting until the profit was sufficiently high to justify paying the fixed cost. Since the profit
was increasing deterministically over time, the firm could pinpoint the particular period of time to enter.
When the profit opportunity is subject to uncertainty, the firm’s objective is the same: wait until the profit
gets high enough to justify paying the fixed cost. Since the prospective profit is evolving stochastically,
the firm cannot pinpoint the exact moment in time that the prospective profit will hit a threshold. It
follows that the firm will choose a threshold for prospective profit at which they will invest. Thus, while
3
Note that we only have a solution for T if ρ > g. If this condition does not hold, then the firm will never undertake the
investment. Think about the intuition as to why the firm will never invest.

8
the time at which the option is exercised is not determined, there is a distribution of times consistent
with the firm’s decision.
Again, let x denote the potential profit of the firm. However, now let’s assume that x follows a
geometric Brownian motion with drift
dx
= µdt + σdz (13)
x
where µ > 0 is the expected rate of change, σ > 0 is the conditional standard deviation, and dz
is an increment
√ of a Wiener process (recall from our discussion of Wiener processes that this means
dz = t dt, where t is drawn from a standard normal distribution).
The way to think about the firm’s problem is as follows. The firm’s potential, x, is a state variable
that is moving around randomly over time. The firm has to decide at what level of potential profit. Also,
there is a fixed cost associated with investing, C. Thus, the firm’s decision is to choose at what point to
pay this cost in order to obtain the value of the investment.
We can think of the firm’s problem as a dynamic programming problem. For example, the value of
the firm’s investment opportunity has the following recursive representation:
1
V (xt ) = EV (xt+1 ) (14)
1+r
where r is the real interest rate used to discount the future and E is the mathematical expectations
operator. This could also be written as
1
V (x) = EV (x0 |x)
1+r
Since we assumed that time was continuous, we need to write this Bellman equation in continuous time:
1
V (x, t) = E[V (x0 , t + ∆t|x)] (15)
(1 + r∆t)

Multiply both sides by (1 + r∆t)

(1 + r∆t)V (x, t) = E[V (x0 , t + ∆t|x)]

Re-arranging this expression yields

r∆tV (x, t) = E[V (x0 , t + ∆t|x) − V (x, t)]

Dividing by ∆t yields
1
rV (x, t) = E[V (x0 , t + ∆t|x) − V (x, t)]
∆t
Now, taking the limit as ∆t → 0,
1
rV (x, t) = E(dV ) (16)
dt
where dV = lim∆t→0 [V (x0 , t + ∆t|x) − V (x, t)]. Note that if we assume that the option to invest never
expires, we can ignore the t in the value function. Furthermore, recall that since x follows a Brownian
motion, we cannot differential V as dV = V 0 (x)dx. Instead, we have to use Ito’s Lemma. Doing so, we
can re-write the continuous time Bellman equation above as
 
1 0 1 00
rV (x) = E V (x)dx + V (x)(dx) 2
(17)
dt 2

9
We can now substitute for dx using equation (13). Taking expectations of the term in brackets and
distributing the dt, we now have:
1 2 2 00
σ x V (x) + µxV 0 (x) − rV (x) = 0 (18)
2
This is just a second-order differential equation in V . Fortunately for us, this differential equation has a
known solution. We will use guess and check to verify. Let’s guess that V (x) has a solution of the form
V (x) = Axβ . Plugging this into the differential equation, we can see that this will be a solution if the
following condition holds:  
1 2 2 1 2
σ β + µ− σ β−r =0 (19)
2 2
Since σ, µ, and r are known parameters, this is just a standard quadratic equation that yields two
solutions for β. But if there are two solutions, how does this affect our guess? Well, we started with a
linear homogeneous differential equation. So, if you dust off that old calculus book, you will recall that
when you have multiple solutions, you can write the solution to the differential equation as the linear
combination of the two solutions:
V (x) = A1 xβ1 + A2 xβ2 (20)
where Ai xβi , i = 1, 2 are the homogeneous solutions to the differential equation. Just looking at this
solution doesn’t help us much. But remember V (x) is the value of the option to invest. As a result, we
can impose some boundary conditions on this value function using economic intuition. Once we do that,
we can then present a relatively intuitive way to derive a threshold for investment.
Let’s start with the first boundary condition that we want to impose. Since V (x) is the value of the
option to invest, it should be true that the value of this option becomes worthless as the profit opportunity
becomes arbitrarily small. We can state this mathematically as

lim V (x) = 0 (21)

x→0

So why does this condition matter? Well, let’s look at equation (19). This equation has two solutions, β1
and β2 . Just by looking at the equation we can see that one of these solutions is positive and one of these
solutions is negative. Let β1 denote the positive solution and β2 denote the negative solution. Condition
(21) will hold iff A2 = 0. So now we can reduce our solution to V (x) = A1 xβ1 .
Now, let’s proceed to our second boundary condition. Since V (x) is the value of the option to invest,
think about the incentive of the firm. If the firm invests, they receive a profit x over each interval of time
[t, t + dt]. The present value of this profit over an infinite horizon is x/(r − µ). They also have to pay a
fixed cost I to undertake the investment. Thus, the net benefit is given as:
x
−I
r−µ
If the value of the option to invest is greater than the expected benefit of investment, then the firm should
hold on to the option. If the value of the option is less than or equal to the expected benefit, then the
firm should invest. It follows that the firm will invest at the precise point in which

x∗
V (x∗ ) = −I (22)
r−µ
where x∗ denotes the threshold.

10
 Recall from the previous condition that the value function has a solution V (x) = A1 xβ1 . It follows
from the second condition that, at the threshold,
x∗
A(x∗ )β = −I (23)
r−µ
where I’ve neglected the subscripts on A and β for expositional convenience. Or, solving for A,
 ∗ 
x
A = (x∗ )−β −I (24)
r−µ

Plugging this back into the solution gives us:

β 
x∗
 
x
V (x) = −I (25)
x∗ r−µ

Now, we have a value function with some economic meaning. Let’s define this as:

x β x∗
   
V (x) = × −I (26)
x∗ r−µ
| {z } | {z }
Stochastic Discount Factor Net Benefit of Investment

So here we have the value of the option as the present discounted value of the expected benefit, where
the discount factor is stochastic because it depends on x. This expression, however, also illustrates the
intuition associated with the option exercise decision. For example, when choosing the threshold, x∗ , a
higher threshold implies a higher net benefit of investment. At the same time, on average, it will take
longer for x to hit the threshold if the threshold is high. This implies that the present value of the net
benefit of investment declines. Thus, when choosing the threshold, a firm must optimally balance this
trade-off. To do so, the firm can simply choose the value of x∗ that yields the maximum option value.
The first-order condition for this maximization problem is
 ∗ 
1 x 1
β ∗ −I = (27)
x r−µ r−µ

Solving this expression for x∗ yields:

 
∗ β
x = (r − µ)I (28)
β−1

This is the threshold for investment. Note that the threshold is a function of the cost of investment, but
also is a complicated function of the time series properties of x since β depends on σ and µ.
Finally, let T ∗ denote the time period when the threshold is reached. Then,

T ∗ = inf{t ≥ 0 : x∗ ≥ x} (29)

6.3 An Intuitive Approach

An alternative way to think about the problem facing the firm with the investment opportunity is to think
of firm as facing a fundamental trade-off between the size of the benefit and the timing of the benefit.
The benefit to the firm from investing is x∗ − I. Thus, the higher the threshold for x∗ , the greater the
benefit to the firm. On the other hand, the greater the threshold, the longer the expected waiting time
will be. The firm’s decision is to choose the threshold that optimally balances this trade-off.

11
 Once you know this is true, you don’t need dynamic programming. If you know the discount factor
associated with the particular stochastic process, you can just substitute this in. For example, you know
the net benefit of the investment is x∗ − I. Thus, for T ∗ that satisfies (29), the present discounted value
of this benefit is  ∗ 
−r(T ∗ −t) x
V (x) = e −I (30)
r−µ
∗
Let’s define B(x) := e−r(T −t) as the stochastic discount factor. Suppose that the firm’s rate of time
preference is r. Then the stochastic discount factor should have the following recursive representation:

1 dB
r= E (31)
dt B(x)

Or,
1
rB(x) = EdB
dt
Using Ito’s Lemma,  
1 0 1 00 2
rB(x) = E B (x)dx + B (x)(dx)
dt 2
Using our expression for dx, this can be re-written as
1 2 2 00
σ x B (x) + µxB 0 (x) − rB(x) = 0 (32)
2
Here, again, we have a linear homogeneous differential equation. Like the one above, it has a solution of
the form:
B(x) = Γ1 xβ1 + Γ2 xβ2 (33)
where β1 and β2 solve the same quadratic equation as we showed in the previous section.
We can now impose the following boundary conditions:

lim B(x) = 0 (34)

x→0

B(x = x∗ ) = 1 (35)
It is straightforward to show that when these conditions holds,
 β
x
B(x) = (36)
x∗

7 The Role of Competition and Uncertainty in the Option Theory

Competition and uncertainty seem crucial to the option theory of investment. Why should a firm wait
before adopting a project with a positive net present value? How can a firm afford to wait when there
is competition? In a perfectly competitive market, would the option to invest even have value? In the
absence of uncertainty would a firm ever choose to wait? In this section, we address these questions to
in order to motivate some intuition about what is going on in the real option model.

12
7.1 Uncertainty Isn’t Everything
A common misperception in the option theory of investment is that the result is driven entirely by
uncertainty. This is not the case. Consider, for example, in the deterministic case that T ∗ satisfies:
   
1 ρI
T ∗ = max ln ,0
g (ρ − g)x0

Note that so long as ρI > (ρ − g)x0 , the optimal choice for the firm is to wait. Intuitively, whenever the
fixed cost is large relative to the initial profitability it pays for the firm to wait.

7.2 Competition
In the option theory of investment, the optimal thing for firms to do is to wait for some period of time
before they invest. However, when game theorists look at the decision-making of firms, they think about
the strategic interaction between competing firms. This begs the question, can firms really afford to wait
to invest when the face competition from other firms? Would competition eliminate the value in waiting?
It turns about that competition does affect the value of the option to wait. However, it does not eliminate
the value of the option to wait.4
Consider an environment in which there are n identical firms that all produce a single homogeneous
good. At time t, firm i produces qi (t) units of output. The price of the good is p(t). The inverse demand
function for the firm is given as
p(t) = D[x(t), Q(t)]
where Q(t) is the aggregate quantity of the good, x(t) is a stochastic demand shock, and Dx , −DQ > 0.
The production of all other firms is denoted as Q−i (t), such that Q(t) = qi (t) + Q−i (t).
Each firm faces a decision about whether to incrementally change productive capacity. The cost of
capacity expansion is given as Kdqi , where K is the per unit cost of expansion and dqi is the size of the
increment.
For simplicity, assume that there are no variable costs of production and therefore the profit for firm
i is
πi [x(t), qi (t), Q−i (t)] = qi (t)D[x(t), qi (t) + Q−i (t)]
Suppose that the demand shock follows a geometric Brownian motion:

dx
= µdt + σdz
x
where µ and σ are parameters and dz is an increment of a Wiener process.
Let’s take a moment to note the similarities and differences between this framework and our previous
frameworks. First, the similarly is that the stochastic nature of the the demand shock influences the value
of the investment decision. Second, exercising control entails a cost, Kdqi . The difference between this
framework and the previous framework is that the decision of firm i is this framework is dependent on the
decisions made by other firms. We therefore have to modify this framework to find a Nash equilibrium.
The firm faces a dynamic programming problem just like in the previous examples. However, they
now have to choose an optimal threshold for x∗ in which they want to expand given their expectation of
what other firms are going to do. However, this decision will be conditional on the effects of all other
firms wanting to exercise the option to expand as well. Fortunately, as shown by Grenadier (2002), we
can treat the firm as myopic.
4
What follows is an example from Grenadier (2002). Also, see Leahy (1993).

13
 Rather than use dynamic programming, however, we can solve this in the simpler way that we
described above. Let x be a multiplicative shock to the profitability of the firm. Such that the firm’s profit
can be written as
πi = xπ̄i = xqi (j)p(Q)
where p(Q) is an inverse demand function. Note that this can similarly be written as p(qi + Q−i ). Let x∗
denote the threshold value at which the firm wants to make the incremental investment in capacity. Thus,
the additional profit for a firm after the incremental investment is x∗ (∂ π̄i /∂qi )dqi . Let V (x) denote the
value of the firm. The value of the firm can be written as

V (x) = Vp (x) + e−r(T −t) [x∗ (∂ π̄i /∂qi )dqi − Kdqi ]

Here, Vp is the present discounted value of the profit with current capacity if no investment in new
capacity is ever made. It follows that the second term on the right-hand side measures the value of the
option to invest in new capacity. Note that given the assumption that x follows a geometric Brownian
motion, we can re-write this as
 β
x
V (x) = Vp (x) + [x∗ (∂ π̄i /∂qi )dqi − Kdqi ] (37)
x∗

where β > 1. We can now find the optimal threshold by maximizing this expression with respect to x∗ .
The first-order condition is given as

β[x∗ (∂ π̄i /∂qi )dqi − Kdqi ] = x∗ (∂ π̄i /∂qi )dqi (38)

Re-arranging and solving for x∗ yields

β K β K
x∗ = = (39)
β − 1 ∂ π̄i /∂qi β − 1 p(Q) + (Q/n)p0 (Q)

We can now consider what happens to the threshold for demand as the number of firms gets infinitely
large:
β K
lim x∗ =
n→∞ β − 1 p(Q)
Note that this implies that competitive firms have a lower threshold than a monopolist. This is consistent
with our intuition that a higher degree of competition should cause firms to act on their investment
opportunities sooner.

7.3 What Does It Mean for Investments to be Irreversible?

The real option literature often focuses on irreversible investment. Firms choose the optimal threshold
at which to exercise their option to invest given the fixed cost associated with the investment. Once the
option to invest is exercised the firm cannot undo the investment. (In some cases, the firm might be able
to recover some of the fixed costs, but even this is only partially reversible investment.) This differs from
the neoclassical model of Jorgenson (1963) in which investment is completely reversible.
As we have detailed, an important result of the real options approach is that the positive net present
value criteria is shown to be an inadequate guide to investment. Some have suggested that this result
is due to the irreversibility of investment. However, it is possible to show that the NPV criteria remains
incorrect even when investment is completely reversible. 5
5
What follows draws on Davis and Cairnes (2015).

14
 Consider the standard real option model. A firm that has the option to produce at rate q or not
produce at all. If the firm decides to produce, they have to incur a fixed cost. Suppose that the price of
the good they are producing is p and follows a geometric Brownian motion:

dp
= µdt + σdz
p
where µ is a drift parameter, σ is a volatility parameter, and dz is an increment of a Wiener process. In
addition assume that µ ∈ (0, r), where r is the real rate of interest.
If the firm is already producing and decides to stop producing, they incur a cost. The cost can be
either positive or negative. If the cost is positive, then there is some dismantling cost. If the cost is
negative, there is some salvage value in exiting.
In this context, the firms investment decision can be understood as the decision to maximize the
expected present value of switching between producing and not.
Davis and Cairnes (2015) make the following modification to the standard model. First, there are no
operating costs. Second, they assume that the cost of opening is k and that the cost of closing is −φk,
where φ ∈ [0, 1]. This assumption is nice because it allows for a comparison of what happens when the
investment is completely irreversible φ = 0 and when investment is completely reversible, φ = 1.
Let V0 (p) denote the value of the project for a firm that is not producing and V1 (p) denote the value
of the project for a firm that is producing. For the firm that is not producing, the value of the project is
solely in the option to invest. Thus, using the methodology we have employed above, the value function
has a solution
V0 (p) = A1 pβ1 + A2 pβ2
For the firm that is already producing, the value of the project is the present value of the profits that the
firm will earn if they never stop producing plus the value of the option to stop producing. Again, using
the methodology described above, the value function has a solution

V1 (p) = Vp (p) + B1 pβ1 + B2 pβ2

where Vp is the particular solution that measures the value of the project if the firm never disinvests.
Note that the following boundary conditions must be satisfied:

lim V0 (p) = 0
p→0

lim V1 (p) = 0
p→∞

Assuming that β1 > 0 and β2 < 0. It follows that A2 = B1 = 0. Let pe denote the price threshold
for entry and px the price threshold for exit. The value-matching and smooth-pasting conditions can be
written as
V0 (pe ) = V1 (pe ) − k (40)
V0 (px ) + φk = V1 (px ) (41)
V00 (pe ) = V10 (pe ) (42)
V00 (px ) = V10 (px ) (43)
Note that for φ < 1, we need to solve a system of 4 non-linear equations to determine four unknowns
(pe , px , A1 , B2 ). Consider two possible scenarios. First, in the case in which investment is completely

15
irreversible, then φ = 0. In this case, it is easy to show that the firm will exit when px = 0. The entry
price is determined through the same process we used above and therefore

β1
pe = k
β1 − 1
However, when φ = 1, it is straightforward to see that pe = px . In other words, in the case in which
investment is completely reversible, the price that determines entry and exit is the same. Brennan and
Schwartz (1985) provide an algorithm for a solution for φ ∈ (0, 1). Taking the limit of their solution as
φ → 1 yields
pe = rk

8 Agency Costs
One issue that has been left out of the discussion of investment thus far is the role of financing. In
a idealized market, there would be no difference between the cost of internal and external finance.
For example, a firm that retained earnings to finance an investment would face an opportunity cost of
investing in the form of foregone interest. A firm that borrowed to finance investment would pay interest
to the lender. In a perfectly competitive market without any sort of financial frictions, these costs would
be identical.
In reality, however, firms might face higher costs of external financing. The reason is due to infor-
mation asymmetries between borrower and lender. A borrower has unique knowledge about the project
that they are investing in and the likelihood of success. The lender, however, needs to have some way
of monitoring the project’s successfulness. In this section, we outline a model that incorporates agency
costs into the investment decision. We then discuss implications for empirical work and some findings in
the literature.

8.1 The Model

Suppose that a firm has to opportunity to undertake a project. The firm’s net worth is W and the cost
of the project is normalized to 1. Assume that W < 1. In order to undertake the project, the firm must
borrow an amount 1 − W . Suppose that there is uncertainty about the value of the project. For example,
suppose that the output of the project is drawn from a uniform distribution on the interval [0, 2γ]. It
follows that the expected value of the project is γ.
A firm that does not undertake the project can invest its net worth and receive a rate of return r.
Assume that the firm is risk-neutral such that it will undertake the project if the expected benefits exceed
the costs.
The important assumption for the model is that the firm can costlessly observe the outcome of its
own project, but the lender cannot. As a result, the lender must pay a cost 0 < c < γ to observe the
outcome of the project.
Let y ∈ [0, 2γ] denote the output of the firm. The firm and the lender will sign a contract that takes
into account the expected return and the potential costs of verification. This contract will include some
critical value of the investment project, D, such that when y > D, the firm pays D to the lender and
when y ≤ D, the lender pays the monitoring cost and seizes the output y. To figure out the implications
of this sort of contract for investment, we need to first determine the equilibrium value of D and then
solve for the equilibrium value of investment.
Consider the perspective of the lender for the case in which D < 2γ. If the firm produces an amount
greater than D, then the lender receives D. The probability that this happens is (2γ − D)/2γ. If the

16
amount is less than D, then the lender receives the output and pays the verification cost. The expected
output is given as D/2. It follows that expected net revenues for the lender are
 
2γ − D D D
E(R) = D+ −c
2γ 2γ 2

If D ≥ 2γ, then the lender receives

E(R) = γ − c
At the point at which the lender is indifferent between borrowing and lending, it will be true that

E(R) = (1 + r)(1 − W )

where it must be true that (1 + r)(1 − W ) ≤ max[E(R)].

It follows that D∗ will satisfy:

D∗ = 2γ − c − (2γ − c)2 − 4γ(1 + r)(1 − W )

p

For the firm, it must be true that the expected output less the borrowing costs be greater than the
benefit from investing in the risk free asset. Formally, this means that

γ − (1 + r)(1 − W ) − A(c, r, W, γ) > (1 + r)W

Note that the borrowing costs include the agency costs associated with borrowing. We can define the
agency costs as
D∗
A= c
2γ
Given the definition of D∗ , it is straightforward to show that

Ac > 0

Ar > 0
AW < 0
Aγ < 0
In other words, agency costs are increasing the size of the monitoring cost and the risk free interest
rate and decreasing in the net worth of the firm and the expected benefit. This leads to the following
implications:

• Agency costs reduce investment for a given interest rate.

• An increase in interest rates reduces the demand for investment because it raises the cost of
borrowing, but also because it increases agency costs.

• A firm’s net worth matters for investment.

• The “financial accelerator”: an increase in interest rates

• The financial system can have an important impact on investment.

17
8.2 Evidence
The agency cost literature suggests that the cost of internal finance is less than the cost of external
finance. One way to test this theory is to examine whether firms that face higher costs of external finance
behave differently. A simple way to test this is to see whether increases in cash flow lead to increases
in investment with the idea being that firms with higher cash flow are more likely to finance internally.
However, this test is too simple. Cash flow is likely correlated with investment for reasons that have
nothing to do with the cost of external finance.
Fazzari, Hubbard, and Peterson (1988) came up with the following identification scheme. Suppose
that one could identify ex ante the firms that were likely to face higher costs of external finance. If so, one
could estimate the effects of cash flow on investment for each group separately. The difference would then
indicate the extent to which external finance matters for investment. To perform this empirical analysis,
they separated firms into groups. Firms that pay high dividends relative to their income were assumed
to be less sensitive to the costs of external finance since they could finance investment by reducing their
dividend. Firms that pay low dividends as a percentage of their income would therefore be more likely
to be sensitive to the costs of external finance.
The authors collected data on individual firms and then separated them in groups. The low dividend
group was defined as firms that paid less than 10% of their income in dividends. The high dividend group
was defined as firms that paid more than 20% of their income in dividends. Their point estimates suggest
that a $1 increase in cash flow increase investment by $0.23 at the high dividend firms and $0.46 at low
dividend firms. A test of the null hypothesis that these coefficient estimates are equal can be rejected at
the 1% level. Thus, their estimates suggest that nearly half of every dollar of cash flow at the low dividend
firms goes to investment, which is twice that of the high dividend firms. They therefore view this as
evidence in favor of the agency cost view.
Others who have found empirical evidence that agency costs matter include ...
Nonetheless, others have been critical of Fazzari et al.’s approach. For example, Kaplan and Zingales
(1997) argue that it is unclear whether their should be a different relationship between cash flow and
investment for low dividend firms. For example, consider a simple model. Suppose that the firm’s
objective is to maximize the net value of the firm. Let the net value be given as

F (I) − rW − C(I − W )

where F (I) is the gross value of the firm with investment I, W is the firm’s net worth and therefore
rW is the opportunity cost of investing, and C(I − W ) is the cost of investing as a function of how
much is borrowed. In addition, assume that C 0 > r and F 0 , −F 00 , C 00 > 0. The first-order condition for
maximization is
F 0 (I) = C 0 (I − W )
In other words, the marginal value of an investment is equal to its marginal cost.
Differentiate this expression with respect to W to obtain
 
dI dI
F 00 (I) = C 00 (I − W ) −1
dW dW
Re-arranging this implies that
dI C 00 (I − W )
= 00 >0
dW C (I − W ) − F 00 (I)
Thus, the theory implies that investment is increasing in the net worth of the firm. What Kaplan and
Zingales argue that Fazzari et al. are doing is testing a hypothesis regarding d2 I/dW 2 , which requires
an assumption regarding C 000 that is not made formally in the literature.

18
9 Some (Largely) Forgotten Literature
An older literature on capital investment focused on something called the “time structure of production.”6
This view of capital investment view capital as “goods-in-process.” The basic idea is that there is an
element of time in the production process. To illustrate this idea, suppose that there is a firm that
produces lumber. At the beginning of the production process, a tree is planted in the ground. Ultimately,
the tree will be cut down and turned into lumber. The firm’s investment decision is to choose the amount
of time between when the tree is planted and when the tree is cut down and turned into lumber.
Within the Bohm-Bawerk/Wicksell framework, production per year can be thought of as a function
of the period of production. Thus, write the production function as
y
= f (θ)
l
where y is output, l is labor, θ is the production period, and f (θ) is the production such that f 0 , −f 00 > 0.
As we said above, Bohm-Bawerk defined capital as a “form of a time-phased collection of growing
consumables” (Hirshleifer, 1970). So what does that mean. Well, imagine our example with trees. We
could think of land with trees growing on it. Suppose that the age of these trees are evenly distributed
across the land. These trees are goods-in-process in the sense that each of the trees will eventually be
made into lumber. The trees near “maturity” will be cut down and turned into lumber. New trees will
be planted and will turn into lumber θ periods from now. Thus, if the trees are evenly distributed by
age, then each period, some trees will mature and be turned into lumber, some trees will be planted, and
the remaining trees will age. In a stationary economy, the trees will remain evenly distributed across age
groups.
We will follow Bohm-Bawerk (and Wicksell and Hirshleifer) and assume that we are in a stationary
economy. This avoids all the trouble that comes with figuring out the transition path from one age
distribution to another.
In keeping with the theme of capital being goods in process, let’s assume that a tree was planted θ
periods ago. Suppose that the only input cost that went into making the tree is the labor that was used
to plant the tree. It follows that the value of any tree is just the present discounted value of the labor
expenditure. Since the tree was planted θ periods ago, we can write this as
Z 0
K= e−rt wldt
−θ

where K is real units of capital, w is the real wage, r is the real interest rate used to discount the future,
and l is the quantity of labor. Note that in a stationary economy, w and l do not depend on t. Thus, this
could be re-written as Z 0
K
=w e−rt dt
l −θ
Integration implies:
K w
= (erθ − 1)
l r
Now define the profit for the firm as
Profit = y − rK − wl
Note that we could write the profit equation in terms of labor:
Profit y K
= −r
l l l
6
See Bohm-Bawerk (1891) and Wicksell (1954). The model used in this section is that of Hirshleifer (1970).

19
Doing so is beneficial since the profit per unit of labor is simply a function of θ. For example, we can
re-write the equation as
Profit
= f (θ) − w(erθ − 1)
l
Since the firm takes the real interest rate and the wage as given, this implies that the profit-maximization
condition is given as
f 0 (θ) = rwerθ
Now, let’s assume that firms are perfectly competitive and earn zero profits. This implies that

f (θ) = werθ

Substituting this into the first-order condition yields

f 0 (θ) = rf (θ)

Or
f 0 (θ) dy
r= = y
f (θ) dθ
This equilibrium condition implies that the profit-maximizing condition for capital (when defined as
goods-in-process) is to set the period of production such that the time rate of growth in production is
equal to the real interest rate. Put differently, if the tree is growing faster than the real interest rate, then
the marginal benefit from leaving the tree in the ground is larger than the marginal cost. One should
leave the tree in the ground. Since we assumed that f 0 , −f 00 > 0, the tree is growing at a decreasing rate
over time. Thus, once the growth of the tree falls such that it is equal to the real interest rate, then the
tree should be cut down and turned into lumber.

References
[1] Abel, Andrew B, Avinash K. Dixit, Janice C. Eberly, and Robert S. Pindyck (1996) “Options, the Value
of Capital, and Investment.” Quarterly Journal of Economics, Vol. 111, No. 3, p. 753 - 777.

[2] Bernanke, Ben S. (1983) “Irreversibility, Uncertainty, and Cyclical Investment.” Quarterly Journal of
Economics, Vol. 98, No. 1, p. 85 - 106.

[3] Clark, John M. (1917) “Business Acceleration and the Law of Demand: A Technical Factor in Economic
Cycles.” Journal of Political Economy, Vol. 25, p. 217 - 235.

[4] Clark, John M. (1944) “Additional Notes on Business Acceleration and the Law of Demand.” Readings
in Business Cycle Theory. p. 254 - 260.

[5] Davis, Graham A. and Robert D. Cairnes (2015) “The Odd Notion of Reversible Investment”, working
paper.

[6] Dixit, Avinash K. (1989) “Entry and Exit Decisions Under Uncertainty.” Journal of Political Economy,
Vol. 97, No. 3, p. 620 - 638.

[7] Dixit, Avinash K. and Robert S. Pindyck (1994) Investment Under Uncertainty. Princeton: Princeton
University Press.

[8] Doms, Mark and Timothy Dunne. 1998. “Capital Adjustment Patterns in Manufacturing Plants.”
Review of Economic Dynamics, Vol. 1, p. 409 - 429.

20
[9] Grenadier, Steven R. (2002) “Option Exercise Games: An Application to the Equilibrium Investment
Strategies of Firms.” Review of Financial Studies, Vol. 15, No. 3, p. 691 - 721.

[10] Hall, Robert E. and Dale W. Jorgenson (1967) “Tax Policy and Investment Behavior.” American Eco-
nomic Review, Vol. 57, p. 391 - 414

[11] Hayashi, Fumio (1982) “Tobin’s Marginal q and Average q: A Neoclassical Interpretation.”

[12] Jorgenson, Dale W. (1963) “Capital Theory and Investment Behavior.” American Economic Review,
Vol. 52, No. 2, p. 247 - 259.

[13] Koyck, L.M. (1954) Distributed Lags and Investment Analysis. North Holland: Amsterdam.

[14] Leahy, John V. (1993) “Investment in Competitive Equilibrium: The Optimality of Myopic Behavior.”
Quarterly Journal of Economics, Vol. 108, No. 4, p. 1105 - 1133.

[15] McDonald, Robert and Daniel Siegel (1986) “The Value of Waiting to Invest.” Quarterly Journal of
Economics, Vol. 101, No. 4, p. 707 - 728.

[16] Tobin, James (1969) “A General Equilibrium Approach to Monetary Theory.” Journal of Money, Credit
and Banking, Vol. 1, No. 1, p. 15 - 29.

21
 Real Business Cycle Notes

Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
We have spent much of the class thus far on partial equilibrium models of consumption, investment, and
asset prices. We now want to venture into general equilibrium models so that we can discuss things like
business cycles and growth. Our first step in this process will be to work through real business cycle
models. We begin with these models for two reasons. First, real business cycle models represent the
core model of modern business cycle research. In other words, most business cycle models that you will
encounter in the literature have the real business cycle model as a special case. Second, the model is
simple (relatively speaking) because it assumes that we live in a world of complete information, rational
expectations, and perfect competition. There are no frictions, or inefficiencies, in the market. This
feature is important because it means that all fluctuations in the business cycle are Pareto efficient. There
is no role for policy.
Based on this description you might think that the RBC model seems odd. Most business cycle
theories rely on some type of inefficiency, like imperfect information or imperfect competition, monetary
policy, fiscal policy, asset bubbles, or shocks to trade. None of these things are in the model. So what
causes fluctuations in an RBC model? The answer is changes in tastes and technology. Changes in
production possibilities or preferences cause individuals and firms to adjust their behavior. The term
“business cycle” just refers to the general patterns of co-movement between output and other economic
variables. Thus, the purpose of the RBC model is to see how well a frictionless model can explain these
co-movements. Or, put differently, how well can the model replicate the true data generating process of
the real world economy.
The RBC model is useful for two main reasons. First, the model serves as a benchmark to compare all
other models. For example, suppose that you think that a certain friction in the economy is important for
explaining the business cycle. If your model doesn’t perform any better than the RBC model at matching
the data, then perhaps that friction isn’t really that useful at explaining the business cycle after all. In
fact, this is what Long and Plosser (1983, p. 68) argued was the purpose of their model:
Although equilibrium real business cycle models of the type we suggest are capable of gen-
erating business-cycle-like behavior, we do not claim to have isolated the only explanation
for fluctuations in real activity. We do believe, however, that models of this type provide a
useful, well-defined benchmark for evaluating the importance of other factors (e.g., monetary
disturbances) in actual business-cycle episodes.
The second reason that the model is important is that it might push us toward a better understanding
of what might cause changes in productivity over time. You will sometime hear people deride the RBC
model because it relies on technology shocks to generate the business cycle. You might hear complaints
that “we don’t get recessions because people forget how to use technology.” I prefer to see the model as a
reduced form model of the business cycle. What the RBC model is really saying is that fluctuations in real

1
GDP tend to be driven by things other than the factors of production.1 The question is therefore about
the source of these fluctuations. Are they really driven by changes in total factor productivity? If so,
why? One possibility is fluctuations in total factor productivity are driven by the reallocation of resources
from one industry to another. When resources flow out of high productivity industries and into low
productivity industries, this lowers aggregate productivity. Of course, this raises the question as to why
this is the case. One possibility is that preference shocks causes demand-side substitutions. In this case,
fluctuations are still efficient. However, an alternative hypothesis is that there are inefficiencies that cause
these substitutions. In that case, we need to modify the model to generate some policy implications.
In what follows, we will use a textbook RBC model in which the only shock is a shock to technology.
The basic setup will be familiar, given our discussion of consumption and investment. The model will
give us a system of nonlinear difference equations that include rational expectations of future outcomes.
As you should know by now, since the term rational expectations means model-consistent expectations,
we cannot treat these expectations as endogenous. As a result, we need to solve our system of equations
to eliminate these expectational terms. These notes will take you through the model and through the
solution process. We will conclude by showing how to solve the model using MATLAB.

2 The Model
Time is discrete and continues forever. The model consists of a representative household and a repre-
sentative perfectly competitive firm. There is no government and there are no frictions in the economy.

2.1 Household
The household owns capital and rents it to firms at a real rental rate, rt . The household also supplies
labor to the firm for a real wage, wt . The income earned from capital and labor are used by the household
to either consume the good produced by the firm or to invest in new capital. The household’s capital
stock evolves according to:
Kt+1 = (1 − δ)Kt + It (1)
where Kt is the capital owned at time t, It is gross investment, and δ is the rate of depreciation. The
household’s budget constraint is
wt Lt + rt Kt = Ct + It (2)
where wt is the real wage, Lt is hours worked, and Ct is consumption.
The household’s expected lifetime utility is
∞
Ct1−σ L1+
X  
t
E0 β − t
1−σ 1+
t=0

Combining the capital accumulation equation with the budget constraint and solving for Kt+1 yields:

Kt+1 = wt Lt + (1 + rt − δ)Kt − Ct (3)

The Bellman equation for the household is

Ct1−σ L1+
V (Kt ) = max − t + βEt V (Kt+1 ) (4)
Ct ,Lt 1−σ 1+
1
This isn’t always the case, as some RBC theorists would attest.

2
Substituting in the budget constraint:

Ct1−σ L1+
V (Kt ) = max − t + βEt V [wt Lt + (1 + rt − δ)Kt − Ct ]
Ct ,Lt 1−σ 1+

The first-order conditions are

Ct−σ = βEt V 0 (Kt+1 ) (5)
Lt = wt βEt V 0 (Kt+1 ) (6)
Combining these conditions yields:
Lt = wt Ct−σ (7)
Also, from the Bellman equation, we have:

V 0 (Kt ) = (1 + rt − δ)βEt V 0 (Kt+1 )

Using the first-order condition for consumption, we can write this as

V 0 (Kt ) = (1 + rt − δ)Ct−σ

Iterating forward yields:

V 0 (Kt+1 ) = (1 + rt+1 − δ)Ct+1
σ

Now, plugging this back into the first-order condition yields:

Ct−σ = βEt (1 + rt+1 − δ)Ct+1

−σ
(8)

2.2 The Representative Firm

The firm’s production function is given as

Yt = zt Ktα L1−α
t (9)

where Yt is total production, α ∈ (0, 1) is a parameter, and zt is total factor productivity and

ln zt = ρ ln zt−1 + εt (10)

where εt is a random shock to productivity.

The firm hires labor and rents capital to produce. The firm’s objective is therefore:

max zt Ktα Lt1−α − wt Lt − rt Kt (11)

Kt ,Lt

The maximization conditions are:

wt = (1 − α)zt Ktα L−α
t (12)
rt = αzt Ktα−1 L1−α
t (13)

3
2.3 Equilibrium
In equilibrium, all of the maximization conditions must hold. Thus, we have

Ct−σ = βEt (1 + rt+1 − δ)Ct+1

−σ
(14)

Lt = wt Ct−σ (15)

wt = (1 − α)zt Ktα L−α
t (16)
rt = αzt Ktα−1 L1−α
t (17)
The goods market must clear:
Yt = Ct + It (18)
Output is given as
Yt = zt Ktα L1−α
t (19)
And productivity is
ln zt = ρ ln zt−1 + εt (20)
Finally, the capital accumulation equation is

Kt+1 = (1 − δ)Kt + It (21)

We now have 8 equations with 8 unknowns: Ct , rt , Lt , wt , Kt , zt , Yt , and It .

We now need to solve these equations for a rational expectations equilibrium. Note that the difficulty
in solving for the rational expectations equilibrium is that we have expectations terms in the model. The
idea of rational expectations is that the expectations formed by the individuals in the model are “model
consistent.” To get a rational expectations solution, we need to solve for each variable in terms of the
state variables in the model. A second issue we have is that we have a system of non-linear equations.
To simplify things, we will begin by log-linearizing around the steady state. Once we have done so, we
can demonstrate how to solve the model.

2.4 Steady State

Solving the steady state of our model. In the steady state a variable, yt = y, ∀t. First, note that from our
time series process for productivity, we have a steady state value of z = 1. Let’s re-write our remaining
equations without the t subscripts:
C −σ = β(1 + r − δ)C −σ
L = wC −σ
w = (1 − α)K α L−α
r = αK α−1 L1−α
Y =C +I
Y = K α L1−α
K = (1 − δ)K + I
From the first equation, we know have
1
r= −1+δ (22)
β

4
Note that we can re-write the firm’s conditions as
 α−1
K
r=α
L

This gives us
 1/(α−1)
K r
=
L α
Plugging this into our equation for w yields:
 α
K
w = (1 − α) (23)
L

So, now we have our steady state factor prices.

From the production function, we can write
 α
Y K
=
L L

Now return to the market clearing condition, we can write:

C Y I
= −
L L L
From the capital accumulation equation, we have I = δK. So we can write:

C Y K
= −δ
L L L
Now, go to the labor supply curve to write
 −σ
C
L+σ = w
L

Thus,
  −σ 1/(+σ)
C
L= w (24)
L
We can then get:
K
K= L (25)
L
C
C= L (26)
L
Y
Y = L (27)
L
I =Y −C (28)
And,
z=1 (29)
These are our steady state values.

5
2.5 Log-linearization
We can log-linearize each of our 8 equations around the steady state. To do so, consider a first-order
Taylor series approximation of a function f (x, y) = 0 around the steady state x = x̄ and y = ȳ:

f (x, y) ≈ f (x̄, ȳ) + fx (x̄, ȳ)(x − x̄) + fy (x̄, ȳ)(y − ȳ)

Since f (x) = 0, then

f (x, y) ≈ +fx (x̄, ȳ)(x − x̄) + fy (x̄, ȳ)(y − ȳ)
Or,
(x − x̄) (y − ȳ)
f (x, y) ≈ +fx (x̄, ȳ)x̄ + fy (x̄, ȳ)ȳ
x̄ ȳ
Note that
(x − x̄)
ln x − ln x̄ ≈ ≡ x̂
x̄
We can therefore write:
f (x, y) ≈ +fx (x̄, ȳ)x̄x̂ + fy (x̄, ȳ)ȳ ŷ
We can now apply this concept to our equations. Consider the consumption Euler equation:

−σC −σ−1 C Ĉt = −σβ(1 + r − δ)C −σ−1 CEt Ĉt+1 + C −σ rEt rt+1

where the variables without a subscript are the steady state values. We can re-write this:

−σC −σ Ĉt = −σβ(1 + r − δ)C −σ Et Ĉt+1 + C −σ rEt rt+1

Dividing through by −σC −σ yields

1
Ĉt = β(1 + r − δ)Et Ĉt+1 − βrEt rt+1 (30)
σ
From the labor supply equation:
L̂t = ŵt − σ Ĉt (31)
From the firm’s first-order conditions:

ŵt = ẑt + αK̂t − αL̂t (32)

r̂t = ẑt + (α − 1)K̂t + (1 − α)L̂t (33)

ẑt = ρẑt−1 + ut (34)
Ŷt = ẑt + αK̂t + (1 − α)L̂t (35)
From the market clearing condition:
C I
Ŷt = Ct + I t (36)
Y Y
Finally, from the capital accumulation equation:

I ˆ
K̂t+1 = (1 − δ)K̂t + It (37)
K

6
2.6 Solving the linearized system
Note that we have three types of variables in our linearized system. We have state variables (those known
at time t), we have “jump” variables (forward-looking variables, or those for which there are expectations),
and we have static variables (those that only appear with a time t subscript). To solve for the rational
expectations equilibrium, we want to write all of the variables in terms of a state variable. We could use
system reduction, whereby we come up with the same number of equations as the sum of jump variables
and state variables. Alternatively, we could write the system of equations in the following form:

AEt Xt+1 = BXt

where X is a vector of the variables in the model. We want to order the variables in a particular way.
Suppose that we put the state variables first, the state variables second, and the jump variables third. We
can now consider how to solve the model.

2.6.1 System Reduction, the Messy Way

As a general rule, there is no reason to believe that A will be invertible. As an alternative, we can use
system reduction methods. To see this, recall our system of equations:
1
Ĉt = β(1 + r − δ)Et Ĉt+1 − βrEt rt+1 (38)
σ

L̂t = ŵt − σ Ĉt (39)

ŵt = ẑt + αK̂t − αL̂t (40)
r̂t = ẑt + (α − 1)K̂t + (1 − α)L̂t (41)
ẑt = ρẑt−1 + ut (42)
Ŷt = ẑt + αK̂t + (1 − α)L̂t (43)
C I
Ŷt = Ct + I t (44)
Y Y
I ˆ
K̂t+1 = (1 − δ)K̂t + It (45)
K
Note from this system that one of our jump variables rt appears in our consumption equation, but does
not appear in any other difference equation. We can use this fact to simplify our model. Let’s consider a
subset of equations:
ŵt = ẑt + αK̂t − αL̂t (46)
L̂t = ŵt − σ Ĉt (47)
r̂t = ẑt + (α − 1)K̂t + (1 − α)L̂t (48)
Combining the first two equations, we have:

L̂t + σ Ĉt = ẑt + αK̂t − αL̂t (49)

Solve this for hours worked to get:

1 α σ
L̂t = ẑt + K̂t − Ĉt (50)
+α +α +α

7
Now plug this into the first-order condition for capital to get:

1+ (α − 1)( + α) + α(1 − α) σ(1 − α)

r̂t = ẑt + K̂t − Ĉt
+α +α +α
Thus, if we collect terms and take expectations, we can eliminate rt , wt , and Lt from our system of
equations. We would then have a system of 5 equations with 5 unknowns (C, z, K , I, Y):
 
σ(1 − α) 1+ (α − 1)( + α) + α(1 − α)
Ĉt = β(1 − r − δ) − Et Ĉt+1 + ẑt+1 + K̂t+1 (51)
+α +α +α

ẑt = ρẑt−1 + ut (52)

I ˆ
K̂t+1 = (1 − δ)K̂t + It (53)
K
C I
Ŷt = Ct + It (54)
Y Y
 
1 α σ
Ŷt = ẑt + αK̂t + (1 − α) ẑt + K̂t − Ĉt (55)
+α +α +α
We could reduce these even further by eliminating the static variables Ŷ and I.
ˆ But this way of reducing
our system is quite tedious and it’s easy to misplace a parameter or two when doing so. So what we
want is a method of system reduction that is more straightforward. So what we need to do is identify
what variables can be reduced from our system. The answer to this question is any variable that can
be written as a linear combination of the jump variables and the state variables. This is all of the static
variables. Furthermore, in our model, as we just demonstrated, this includes r̂ as well. So we can include
this among our static variables. In addition, as a general rule, one can include jump variables that do
not have their “own” difference equation (e.g., Et r̂t+1 appears in the consumption Euler equation and
nowhere else; there is no equation with both Et r̂t+1 and rt ).

2.6.2 System Reduction, the Cleaner Way

Recall that we wrote our system of linearized equations as

AEt Xt+1 = BXt

Consider writing out our system of equations like this:

     
a11 a12 X1,t+1 b11 b12 X1,t
Et =
a21 a22 X2,t+1 b21 b22 X2,t

where X1,t is a vector of the jump and state variables and X2,t is a vector of all variables that can
be solved as a linear combination of X1,t . Before proceeding, let’s consider the dimensions of these
sub-matrices. If there are n difference equations in the model, then assume that a11 and a12 have n
rows each. This assumption implies that a21 and a22 are matrices of zeros. Suppose there are N total
variables in the model. We need to be able to compute the product a11 X1,t+1 . This requires that the
number of columns in a11 must be the same as the number of jump/state variables. A proper system
reduction should reduce the system into the same number of variables as there are difference equations.
This implies that
a11 : n × n
a12 : n × (N − n)

8
 a21 : (N − n) × n
a22 : (N − n) × (N − n)
And correspondingly for B. Now let’s re-write our system:
     
a11 a12 X1,t+1 b11 b12 X1,t
Et =
0 0 X2,t+1 b21 b22 X2,t

We can therefore write

0 = b21 X1,t + b22 X2,t
If b22 is invertible, we get
X2,t = −b−1
22 b21 X1,t

So the static variables can be written as a function of the jump variables and state variables. Now re-write
the system as
     
a11 a12 X1,t+1 b11 b12 X1,t
Et =
0 0 −b−1
22 b21 X1,t+1 b21 b22 −b−1
22 b21 X1,t

From the first equation we now have:

   
−1 −1
Et a11 − a12 b22 b21 X1,t+1 = b11 − b12 b22 b21 X1,t

Or,
CEt X1,t+1 = DX1,t

2.6.3 Working with the Reduced System

We showed how to get our system of equations reduced down to something more manageable. Since
matrix C is square and of full rank, it is invertible. So we can re-write our system as

Et X1,t+1 = M X1,t (56)

From the definition of eigenvalues, we can decompose M as

M = ΓΛΓ−1

where Γ is a matrix of eigenvectors and Λ is a diagonal matrix of eigenvalues. We can therefore write
our system as
Et X1,t+1 = ΓΛΓ−1 X1,t (57)
Now order the eigenvalues from largest to smallest (I’m having you do it this way because MATLAB will
do it this way for you). Pre-multiply both sides by Γ−1 and define yt := Γ−1 X1,t to get

Et yt+1 = Λyt (58)

Let ` denote the number of unstable eigenvalues and p denote the number of stable eigenvalues. Since
we have arranged the eigenvalues from largest to smallest, we can write
    
y1,t+1 Λ11 0 y1,t
Et =
y2,t+1 0 Λ22 y2,t

9
where Λ11 is a ` × ` matrix of unstable eigenvalues and Λ22 is a p × p matrix of stable eigenvalues. If this
is true, then the only way that there is a solution is if that y1,t = 0. Otherwise, y1 explodes over time.
Finally, recall that yt = Γ−1 X1,t . It follows that we can write this definition as
   " j #
y1,t G11 G12 X1,t
= s
y2,t G21 G22 X1,t
| {z }
Γ−1

j
where X1,t denotes the jump variables and X1,t
s denotes the state variables. We showed that it must be

true that y1,t = 0. This implies that

j s
0 = G11 X1,t + G12 X1,t

j
Solving for X1,t yields:
j
X1,t = −G−1 s
11 G12 X1,t (59)
For this to be a solution, G11 must be invertible. So G11 must be square. In addition, we must be able to
j
calculate the following products: Λ11 G11 and G11 X1,t . Thus, the number of rows in G11 must be equal
to the number of columns in Λ11 and the number of columns in G11 must be equal to the number of
jump variables. Thus, the dimensions of G11 must be ` × n. Since G11 has to be square, this requires
that ` = n. In other words, to get a unique solution, we need to have the same number of unstable
eigenvalues as jump variables.2
Let’s assume that our uniqueness condition is satisfied, we can now go back to

Et X1,t+1 = M X1,t

We can re-write this as " # " #

j  j
X1,t+1 M11 M12 X1,t
Et s = s
X1,t+1 M21 M22 X1,t
j
Of, course, we already know that X1,t = −G−1
11 G12 X1,t . So we can re-write this as
s

−G−1 s −G−1 s
    
Et 11 G12 X1,t+1 =
M11 M12 11 G12 X1,t
s
X1,t+1 M21 M22 s
X1,t

From the second equation, we have

s
X1,t+1 s
= (M21 χ + M22 )X1,t (60)

where χ = −G−1
11 G12 .
Thus, we now have:
s
X1,t s
= (M21 χ + M22 )X1,t−1 (61)
j
X1,t = −G−1 s
11 G12 X1,t (62)

X2,t = −b−1
22 b21 X1,t (63)
2
If we have too many unstable eigenvalues, then there is no unique equilibrium. If we do not have enough unstable
eigenvalues, we will have an infinite number of possible equilibria.

10
3 Solving the Model
Now let’s use this solution procedure to solve our model. We need to write our model in the following
form:
AXt+1 = BXt
Recall our system of equations, written with the difference equations ordered first:

Et Ĉt+1 − σ1 βrEt rt+1 = Ĉt

K̂t+1 = (1 − δ)K̂t + δ Iˆt
ẑt+1 = ρẑt + ut+t
0 = ŵt − σ Ĉt − L̂t
0 = ẑt + αK̂t − αL̂t − ŵt
0 = ẑt + (α − 1)K̂t + (1 − α)L̂t − r̂t
0 = ẑt + αK̂t + (1 − α)L̂t − Ŷt
C I
0 = Y Ct + Y It − Ŷt

where we have not only moved all the t + 1 terms to the left-hand side and all the t terms to the right-
hand side, but we have simplified the equations since we know from the steady state conditions that
β(1 + r − δ) = 1 and I/K = δ. So,
   
Ĉt+1 Ĉt
 
1 0 0 0 0 0 0 −(1/σ)βr 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0  K̂t+1 0 (1 − δ) 0 0 δ 0 0 0 K̂t
    
     

 0 0 1 0 0 0 0 0 
 ẑt+1  
  0 0 ρ 0 0 0 0 0 
 ẑt 

0 0 0 0 0 0 0 0  Ŷt+1 −σ 0 0 0 0 − 1 0 Ŷt
    
 =
    
0 0 0 0 0 0 0 0  Iˆt+1 0 α 1 0 0 −α −1 0  Iˆt
 
     
0 0 0 0 0 0 0 0  0 α−1 1 0 0 1−α 0 −1
   
L̂t+1 L̂t

     
0 0 0 0 0 0 0 0  0 α 1 −1 0 (1 − α) 0 0
   
  ŵt+1   ŵt 
0 0 0 0 0 0 0 0 r̂t+1 (C/Y ) 0 0 −1 (I/Y ) 0 0 0 r̂t

So,  
1 0 0
a11 = 0 1 0 
0 0 1
 
0 0 0 0 −(1/σ)βr
a12 =  0 0 0 0 0 
0 0 0 0 0
 
1 0 0
b11 =  0 1 − δ 0 
0 0 ρ
 
0 0 0 0 0
b12 =  0 δ 0 0 0 
0 0 0 0 0
 
−σ 0 0

 0 α 1 

b21 = 
 0 α−1 1  
 0 α 1 
(C/Y ) 0 0

11
  
0 0 − 1 0

 0 0 −α −1 0 

b22 =
 0 0 1−α 0 −1 

 −1 0 (1 − α) 0 0 
−1 (I/Y ) 0 0 0
Okay, so we need to solve this. So let’s go to MATLAB. We need to enter everything that we have done
so far. Start with the parameters. We need to figure out how to calibrate the parameters of our model.
So let’s create targets that we want to satisfy. One target is that in the U.S., investment is roughly 15% of
GDP. So we should have
C
= 0.85
Y
We need to know what parameterization makes this approximately true. Recall that we solved for C/L
and Y /L in our steady state conditions above:
C Y  1−α  (1−α)/(α−1) (1−α)/(α−1)
− δK δK δK

L L L L L K r (1/β) − 1 + δ
Y
= Y
= 1− Y
= 1− K α = 1−δ = 1−δ = 1−δ
L L L (L) L α α

So,
 (1−α)/(α−1)
(1/β) − 1 + δ
1−δ = 0.85
α
Or,
 (1−α)/(α−1)
(1/β) − 1 + δ
0.15 = δ
α
Let’s start with what we know. We know that α is the capital share of income. Thus, we set α = 1/3. So,

(1/β) − 1 + δ −1
 
0.15 = δ
(1/3)
 
(1/3)
0.15 = δ
(1/β) − 1 + δ
δ
0.45 =
[(1/β) − 1 + δ]
[(1/β) − 1 + δ]0.45 = δ
[(1/β) − 1]0.45 = 0.55δ
Setting β = 0.99,
δ ≈ 0.0083
In other words, from this condition we have,

β = 0.99

α = 0.33
δ = 0.0083
We will then assume that ρ = 0.95, σ = 1,  = 1.
These assumption imply that the rate of time preference is 4 - 5%, we have log utility for consumption,
we have a unit labor supply elasticity, an annual depreciation rate of 3.3%, and a capital share of income

12
of 33%. We can then calculate the steady state. Enter matrices A and B however you like. Then, calculate
M using
 −1  
−1 −1
M = a11 − a12 b22 b21 b11 − b12 b22 b21

You should get:  

0.9916 −0.0093 0.0250
M =  −0.0756 1.0194 0.0839 
0 0 0.9500
Now, we need to decompose the system into M = ΓΛΓ−1 . To do so, we are going to use the following
MATLAB commands:
• [Gamma,Lambda] = eig(M);

• [Gamma,Lambda] = sortem(Gamma,Lambda);
The first command computes a matrix of eigenvectors and a matrix of eigenvalues that satisfy our
decomposition of M . The second command sorts these so that the eigenvalues are ordered from largest
to smallest and the eigenvectors are sorted correspondingly. If you do this, we can solve for Γ−1 :
 
1.2920 −0.7486 −0.3565
Γ−1 =  −1.4608 −0.3093 −2.4416 
0 0 2.8951
 
1.0354 0 0
Λ= 0 0.9756 0 
0 0 0.9500
We have one unstable eigenvalue and one jump variable (C). Thus, we have a unique rational expectations
equilibrium. Note that this also implies that

G11 = 1.292

So, this is obviously invertible since it is a scalar. In fact, we can write:

j
X1,t = −G−1 s
11 G12 X1,t

Or,  
1   K̂t
Ĉt = − −0.7486 −0.3565
1.292 ẑt
Or,
Ĉt = 0.58K̂t + 0.28ẑt
We can then compute
s
X1,t = (−M21 G−1 s
11 G12 + M22 )X1,t−1

Using the model, we should have:

    
K̂t 0.9756 0.0630 K̂t−1
=
ẑt 0 0.9500 ẑt−1

Or,
K̂t = 0.98K̂t−1 + 0.06ẑt−1

13
 ẑt = 0.95zt−1
Finally, recall from system reduction that we have

X2,t = −b−1
22 b21 X1,t

Or,
   
Ŷt −0.5038 0.4962 1.5038
 Iˆt  
 
   −9.1025 3.3338 10.1025 
 Ĉt
 L̂  = 
 t   −0.7519 0.2481 0.7519 
 K̂t
 
 ŵ  
t 0.2481 0.2481 0.7519  ẑt
r̂t −0.5038 −0.5038 1.5038
We can now do impulse response analysis, simulate data, and see how well our model matches the
moments in real world data. The IRFs for consumption, investment, and output are shown in Figure 1.

Figure 1: Impulse Response Functions

References
[1] Blanchard, Olivier J. and Charles M. Kahn. 1980. “The Solution of Linear Difference Models under
Rational Expectations.” Econometrica, Vol. 48, No. 5, p. 1305 - 1311.

[2] DeJong, David N. and Chetan Dave. 2011. Structural Macroeconometrics, Second Edition. Princeton:
Princeton University Press.

14
[3] King, Robert G. and Mark W. Watson. 2002. “System Reduction and Solution Algorithms for Singu-
lar Linear Difference Systems under Rational Expectations.” Computational Economics, Vol. 20, No.
1 - 2, p. 57 - 86.

[4] Long, John B. and Charles I. Plosser. 1983. “Real Business Cycles.” Journal of Political Economy, Vol.
91, No. 1, p. 39 - 69.

[5] McCandless, George. 2008. The ABCs of RBCs. Cambridge: Harvard University Press.

[6] Sims, Eric. 2015. “Notes on Solving Linearized Rational Expectations Models.”

15
 The New Keynesian Model

Dr. Joshua R. Hendrickson

University of Mississippi

1 Introduction
In our previous lecture, we looked at the Real Business Cycle model. In that model, we had complete
information, perfect competition, and the lack of any sort of friction that would have generated a role
for policy. In this lecture, we would like to discuss the New Keynesian model. The model gets its name
because of the familiar structure of the equilibrium conditions. The equilibrium consists of a dynamic IS
equation, a New Keynesian Phillips curve, and a monetary policy rule. The model therefore resembles
earlier linear, IS-LM-type rational expectations models. But what is different about this model?
The main difference between the New Keynesian model and the RBC model is the introduction of
monopolistic competition. The representative perfectly competition firm a RBC theory is replaced by a
continuum of firms with unit mass who each produce their own unique brand of a consumption good.
Since the firms sell differentiated products, they have some control over the price that they are able to
charge. The standard assumption in New Keynesian models is that prices are slow to adjust. There are
a couple of possible explanations for this. One is that prices are set by contracts and therefore cannot
be changed until the terminal date of the contract. A reduced form approach to this sort of idea is what
is known as Calvo (1983) pricing, in which we assume that only a fraction of firms can change their
prices in any given period. Another explanation for the slow adjustment of prices is the assumption
that price adjustments are costly. In other words, there is some sort of adjustment cost or menu cost
to changing prices and therefore firms would prefer to wait to change prices. One way to model this is
through an (s, S) model in which firms set bounds for action. This is similar to the option approach
to investment. Firms prefer inaction (not changing their price) until they hit some critical threshold at
which it is optimal to adjust their price. An alternative is to assume a quadratic adjustment cost, as
in Rotemberg (1982). We will pursue this approach in the notes. The solution to the firm’s pricing
problem can be used to generate a relationship between aggregate output and inflation. This relationship
resembles an expectations augmented Phillips Curve. New Keynesian prefer to see it as a more updated
version of the traditional Phillips curve with inflation on the left-hand side of the equation and the output
gap on the right-hand side. In reality, this is just an equilibrium condition. Causation isn’t running in one
direction or the other. Regardless, this equation has come to be known as the New Keynesian Phillips
Curve.
The slow adjustment of prices creates an inefficiency. Without this cost, prices would immediately
adjust. The fact that they do not adjust like they should suggests a potential role of policy. Since this
involves prices, one potential policy prescription involves monetary policy. If adjusting prices is costly
and prices do not respond as quickly as they should to exogenous shocks, then one potential policy
prescription is to have zero inflation. This implies that prices, on average, are not changing. This
minimizes the inefficiency associated with costly price adjustments.
Despite the focus on monetary policy, one distinct feature of the New Keynesian model is the absence
of money. Instead monetary policy is conducted solely by changing the short term nominal interest

1
rate. This idea originated with Michael Woodford (1999), who was thinking about how monetary policy
might look in a world in which the demand for cash all but disappeared and in which banks had
reserve balances near zero. While this research was initially about a hypothetical “cashless”, a number
of researchers showed that the model could apply to an economy with money as well. For example, they
showed that by putting money in the utility function could add a money demand curve to the model, but
if the central bank conducted policy solely by changing the interest rate then money was redundant in
the sense that you could solve the standard New Keynesian model without the money demand equation.
What is needed to close the model is a monetary policy rule. We need an equation that can summarize
how the interest rate responds to other economic variables in the model. Given the desirability of zero
inflation, one possibility is that the central bank simply adjusts the nominal interest rate to changes in
the rate of inflation. To know whether this is sufficient, we need some type of formal welfare criteria to
determine what type of rule would be optimal. We will discuss how to derive this welfare criteria in the
context of the model. Nonetheless, we will initially assume that the central bank follows a Taylor-type
rule (Taylor, 1993). John Taylor developed the Taylor rule, in which the central bank adjusts the nominal
interest rate positively to increases in inflation relative to its target and the output gap. The basic idea is
that if inflation is higher than its target and/or real GDP is above potential, this is a sign that monetary
policy is too expansionary and therefore policy needs to tighten. A higher nominal interest rate, because
prices are sticky, leads to a higher real interest rate, which reduces output and inflation.1 Taylor (1999)
later found that his rule not only was desirable in model simulations, but that the rule appeared to
describe monetary policy during the period known as the Great Moderation really well. Taylor has since
used this experience to argue that the Taylor rule is a useful guide for policy. I think that there is reason
for skepticism about these claims. For example, I argue that the success experienced during the Great
Moderation (1985 - 2007) was due to the Fed taking responsibility for inflation rather than following a
Taylor rule (Hendrickson, 2012). David Beckworth and I also challenge the effectiveness of a Taylor-type
rule in practice on the grounds that the output gap is not known in real time. All we have are estimates.
We show that this has important implications for economic fluctuations (Beckworth and Hendrickson
2017). We will discuss some of these issues at the end of the lecture. For now, we will close the model
with a Taylor-type rule.
For an early example of a New Keynesian-type model, see Rotemberg and Woodford (1995) who
amended the RBC model to include imperfect competition. For a textbook treatment of the New Keyne-
sian model, see Woodford (2003) or a much more concise Galí (2008). What follows is a lot of tedious
algebra, but such is the nature of the model.

2 The Model
Time is discrete and infinite. The model consists of households and monopolistically competitive firms.
We will assume that there is an index of firms with unit mass that are indexed by i. Households consume
the goods produced by all of the firms. However, their utility is over aggregate consumption. Firms,
since they are monopolistically competitive have some control over their price. We will assume that firms
face a quadratic cost of adjustment with regards to prices. Finally, there is a central bank that conducts
monetary policy according to a Taylor rule.
1
Or so the story goes. See Cochrane (2011) for skepticism about this view.

2
2.1 Households
The expected lifetime utility for our household is
∞
ct1−σ h1+
X  
t
E0 β − t
1−σ 1+
t=0

where ht is hours worked, σ and  are parameters, and ct is aggregate consumption such that
Z 1 θ/(θ−1)
(θ−1)/θ
ct = ct (i) di
0

where ct (i) is the quantity consumed of the good produced by firm i.

Households can save by purchasing bonds from borrowers and borrow by issuing bonds of their own
(since everyone is identical, these will be in zero net supply in equilibrium). The budget constraint can
be written in nominal terms as
Z 1
Bt+1
pt (i)ct (i)di + = Bt + Dt + Wt ht
0 Rt
R1
where 0 pt (i)ct (i)di is aggregate consumption, Bt is the nominal bond balances, Dt is the dividends
paid by the firms, Rt is the (gross) nominal interest rate, and Wt its he nominal wage. Suppose that there
is a price index Pt such that
Z 1
Pt ct = pt (i)ct (i)di
0
We will later verify that such a price index exists. However, what this implies is that we can write the
budget constraint as
Bt+1 Bt
ct + = + dt + wt ht
Rt Pt Pt
where wt is the real wage and dt is the aggregate real dividend. Define bt = Bt /Pt . We can simplify our
budget constraint further to get
πt+1
bt+1 = bt + wt ht − ct (1)
Rt
where π is the (gross) rate of inflation.
The Bellman equation for the household can be written as:

ct1−σ h1+
 
t
V (bt ) = max − + βEt V (bt+1 )
ct ,ht 1 − σ 1+
or
c1−σ h1+
   
t t Rt
V (bt ) = max − + βEt V (it bt + wt ht − ct )
ct ,ht 1 − σ 1+ πt+1
The first-order conditions are given as

Rt
ct−σ = βEt V 0 (bt+1 )
πt+1

Rt
ht = wt βEt V 0 (bt+1 )
πt+1

3
Combining these two conditions yields:
ht = wt c−σ
t

Or, log-linearizing this equation yields:

ĥt = ŵt − σĉt (2)
Differentiating the value function with respect to bt yields
1
V 0 (bt ) = β Rt V 0 (bt+1 )
πt+1
From the first-order condition for consumption, we can re-write this as

V 0 (bt ) = c−σ
t

Iterating forward and plugging this into the first-order condition for consumption yields:
Rt −σ
c−σ
t = βEt c
πt+1 t+1
Log-linearizing gives us:
R −σ−1 1 R
−σcσ−1 cĉt = −σβ c cEt ĉt+1 + β Rc−σ R̂t − β 2 πc−σ Et π̂t+1
π π π
In the steady state we have:
R
β =1
π
So, the above linearization reduces to:

−σĉt = −σEt ĉt+1 + R̂t − Et π̂t+1

Or,
1
ĉt = Et ĉt+1 −
(R̂t − Et π̂t+1 )
σ
Note that there is no investment in our model. Thus, everything that gets produced by firms must be
consumed. Let aggregate production be given as yt . It follows that
1
ŷt = Et ŷt+1 − (R̂t − Et π̂t+1 ) (3)
σ
We have thus transformed our consumption Euler equation into what the New Keynesians call a dynamic
IS curve, since it relates output with the real interest rate.
Okay, now let’s go back to our consumption problem. We need to prove that
Z 1
Pt ct = pt (i)ct (i)di
0

So let’s get started.

Let Z denote total expenditure. We want to show that there is some price level, Pt , such that
Zt = Pt ct . The objective of the consumer is to maximize consumption. The Lagrangian can be written:
Z 1 θ/(θ−1) Z 1 
(θ−1)/θ
max ct (i) di −λ Pt (i)ct (i)di − Z
ct (i) 0 0

4
The maximization condition is
Z 1 [1/(θ−1)]
−1/θ (θ−1)/θ
ct (i) ct (i) di = λPt (i)
0

Note from the definition of ct that

Z 1 [1/(θ−1)]
1/θ
ct (i)(θ−1)/θ di = ct
0

So we can re-write out maximization condition as

1/θ
ct (i)−1/θ ct = λPt (i)
Solve for Pt (i) and plug this into the expenditure definition:
Z 1
1/θ 1/θ (θ−1)/θ
λZ = ct ct (i)(θ−1)/θ di = ct ct = ct
0

So,
ct
λ=
Z
Plug this into our maximization condition to get
1/θ ct
ct (i)−1/θ ct = Pt (i)
Z
Now, suppose that Z = Pt ct . This condition can be re-written as
Pt (i) −1/θ
ct (i)−1/θ = c
Pt t
Or,
 −θ
Pt (i)
ct (i) = ct (4)
Pt
Now, we have a demand curve for each consumption good as a function of the relative price of the
consumption good and total consumption. Furthermore, note that −θ is the price elasticity of demand.
Now we need to show that Pt is a price level. Return to our definition of aggregate consumption. We
can use this demand curve to re-write this as:
Pt (i) −θ (θ−1)/θ θ/(θ−1)
 Z 1    
ct = ct di
0 Pt
Or,
Z 1 1−θ 
(θ−1)/θ (θ−1)/θ Pt (i)
ct = ct di
0 Pt
Z 1 
1 = Ptθ−1 Pt (i)1−θ di
0
Z 1 1/(1−θ)
1−θ
Pt = Pt (i) di
0
This is our Dixit-Stiglitz price index! So we have verified that a price level exists such that
Z 1
Pt ct = Pt (i)ct (i)di
0

5
2.2 Firms
We are going to think about the firm’s choice in two stages. We will assume that the firm wants to
minimize its labor costs. The firm will then want to choose its price to maximize its profit, given the
demand for its product.
Suppose that each firm has a production function given as

yt (i) = at ht (i)

where at is productivity and is stochastic with a mean of unity. The firm would like to minimize its total
cost wt ht subject to this production function. Let Λt be the Lagrangian multiplier from this minimization
problem. The minimization condition can then be written as

Λt at = wt (5)

Or,
Λt yt (i) = wt ht (i)
Firms pay a real dividend to households each period. We can write this as
 2
Pt (i)yt (i) φp Pt (i)
dt (i) = − wt ht (i) − −1 yt
Pt 2 πPt−1 (i)
 2
φp Pt (i)
where 2 πPt−1 (i) −1 ct is an adjustment cost, or menu cost, associated with changing prices over
time. We measure this cost in proportion to aggregate output. Market clearing requires that

yt (i) = ct (i), ∀i

yt = ct
As shown in equation (4), firms face a demand curve for their product from households. Given the
market-clearing condition, we can re-write the demand curve as
 −θ
Pt (i)
yt (i) = yt
Pt

Substituting this demand curve and the minimization condition into the dividend equation yields:
 1−θ  −θ  2
Pt (i) Pt (i) φp Pt (i)
dt (i) = yt − Λt yt − − 1 yt
Pt Pt 2 πPt−1 (i)

The firm’s problem is to choose Pt (i) to maximize the dividend. The profit-maximization condition is:
   
−θ θ−1 −θ−1 θ φp Pt (i) Pt+1 (i) Pt+1 (i)
(1−θ)Pt (i) Pt yt +θΛt Pt (i) P t yt − −1 yt +βφp Et −1 yt+1 = 0
πPt−1 (i) πPt−1 (i) πPt (i)2 πPt (i)

Divide through by yt to get

   
−θ φp Pt (i) Pt+1 (i) Pt+1 (i)
(1 − θ)Pt (i) Ptθ−1 + θΛt Pt (i)−θ−1 Ptθ − − 1 + βφp Et −1 = 0
πPt−1 (i) πPt−1 (i) πPt (i)2 πPt (i)

6
(Note that there is no growth in out model, so Et yt+1 /yt = 1.) This condition is the same for all firms.
Thus, in a symmetric equilibrium, all firms will choose the same price, Pt (i) = Pt . The profit condition
can now be written as
   
−1 −1 φp πt Pt+1 πt+1
(1 − θ)Pt + θΛt Pt − − 1 + βφp −1 =0
πPt−1 π πPt2 π

Multiply both sides by Pt to get

   
πt πt πt+1 πt+1
(1 − θ) + θΛt − φp − 1 + βφp Et −1 =0
π π π π

Note that in the steady state:

θ−1
Λ=
θ
There is zero inflation in the steady state and therefore π = 1. Thus,

(1 − θ) + θΛt − φp (πt2 − πt ) + βφp Et (πt+1

2
− πt+1 ) = 0

Log-linearization yields:
θΛΛ̂t − φp π̂t + βφp Et π̂t+1 = 0
Or,
θ−1
π̂t = βEt π̂t+1 + Λ̂t
φp
Ideally, we’d like to simplify this further. We can log-linearize equation (5) to get:

Λ̂t = ŵt − ât

From equation (), this can be re-written as

Λ̂t = ĥt + σ ŷt − ât

From the production function, ĥt = ŷt − ât , and therefore

Λ̂t = (ŷt − ât ) + σ ŷt − ât = ( + σ)ŷt − ( + 1)ât

We can now write our log-linearized maximization condition as

θ−1
π̂t = βEt π̂t+1 + [( + σ)ŷt − ( + 1)ât ]
φp

Or,
π̂t = βEt π̂t+1 + κŷt + γât (6)
where
(θ − 1)( + σ)
κ=
φp
(θ − 1)( + 1)
γ=−
φp
This equation is a New Keynesian Phillips Curve.

7
2.3 Central Bank
Finally, there is a central bank. We will assume that the central bank conducts monetary policy using a
Taylor-type rule. Specifically, we will assume that the central bank sets the nominal interest rate according
to
R̂t = φπ π̂t + φy ŷt + et (7)
where φπ and φy are parameters chosen by the central bank and et is a monetary policy shock (a
deviation from the policy rule). What this policy rule implies is that the central bank adjusts the nominal
interest rate in response to fluctuations in output and inflation.

2.4 Equilibrium
Finally, we’ve arrived at an equilibrium after all that tedious algebra. We have a system of 3 equations
with 3 endogenous variables:
1
ŷt = Et ŷt+1 − (Rt − Et πt+1 ) (8)
σ
π̂t = βEt π̂t+1 + κŷt + γât (9)
R̂t = φπ π̂t + φy ŷt + et (10)
We can use system reduction to re-write this system as

Et ŷt+1 + σ1 Et π̂t+1 = (1 + σ1 φy )ŷt + σ1 φπ π̂t + σ1 et

γ
Et π̂t+1 = 1 κ
β π̂t − β ŷt − β ât

Or,
1 (1 + σ1 φy ) 1 1
         
1 σ Et
ŷt+1
= σ φπ ŷt
+ σ 0 et
0 1 π̂t+1 − βκ 1
β π̂t 0 − βγ ât
Or,
AEt Xt+1 = BXt + Cet
We need to find A−1 , so that we can re-write this as

Et Xt+1 = M Xt + Det

where M = A−1 B and D = A−1 C. Fortunately, we have a 2 × 2 matrix, so we can write:

1 − σ1
 
−1
A =
0 1

So,
" φπ
# "
φπ
#
1 − σ1 (1 + σ1 φy ) 1 + σ1 φy + κ 1

σ σβ σ − σβ
M= =
0 1 − βκ β
1
− βκ 1
β

We can now determine the conditions for stability finding the eigenvalues of M . In our previous lecture
on the RBC model, we calibrated the model and showed how to solve. We could do the same thing
here. However, what we want to do is determine what role the policy parameters chosen by the central
bank have on the equilibrium properties of the model. For example, we showed in our previous model
the conditions for a unique rational expectations equilibrium. Is it possible that by choosing the right
monetary policy can guarantee that we end up in a unique equilibrium? Is it possible that by choosing the

8
wrong policy parameters that we could end up with multiple equilibria? These are potentially important
questions.2 To explore this issue, recall that from the definition of eigenvalues, we have
 
 1 + 1 φ + κ − λ φπ − 1 
 σ y σβ σ σβ  = 0
− βκ 1
− λ
 
 β 

Given a parameterization, this would be easy to determine. We’d simply solve a quadratic equation.
However, since we are not going to calibrate the parameters, we need to find some other way. Well,
there are two unique things about eigenvalues. Specifically, the product of two eigenvalues is equal to the
determinant of M and the sum of the two eigenvalues is equal to the trace of M . This implies that
 
1 φy κ κ 1
λ1 λ2 = + + + φπ −
β σβ β 2 σ σβ β

φy κ 1
λ1 + λ2 = 1 + + +
σ σβ β
Note that we have reduced our system of equations to two equations and we have two “jump” variables
and no state variables. So we need to have 2 unstable eigenvalues. This implies that we should have

(λ1 − 1)(λ2 − 1) > 0

Or,
λ1 λ2 + 1 > λ1 + λ2
Plugging in these values and eliminating like terms gives us
 
φy κ κ 1 φy κ
+ 2 + φπ − > +
σβ β σ σβ β σ σβ

Multiply through by βσ  
κ 1
φy + + κ φπ − > βφy + κ
β β
Bring φy and φπ to the left hand side to get:

(1 − β)φy + κφπ > κ

1−β
φy + φπ > 1
κ
Thus, for us to have two unstable eigenvalues, the condition above must hold. Note that a sufficient
condition for this to hold is φπ > 1. This condition is known as the Taylor principle.3 The basic idea is
that to ensure that there is a unique rational expectations equilibrium, the central bank needs to increase
the nominal interest rate by more than the increase in the rate of inflation. The intuition is that since the
nominal interest rate is increasing by more than the rate of inflation, the real interest rate will rise, which
will reduce output and inflation.
2
The importance of these questions go beyond the simple presentation here. In fact, if there are multiple equilibria this
raises a number of questions. Do we believe that multiple equilibria in our model has real world implications? Are multiple
equilibria “learnable”? Is this merely a theoretical possibility detached from reality? Or might this be a Keynesian story of
business cycle fluctuations.
3
Arguably, it should be referred to as the Howitt principle, since economist Peter Howitt had explained this long ago. For a
discussion see Howitt (2005).

9
3 The Taylor Principle and Monetary Policy, In Practice
The Taylor principle is an important conclusion of our model. However, one might wonder whether or
not the Taylor principle is important in practice. Taylor (1999) estimated his rule using U.S. data for
the period from 1960 - 1979 and finds that the coefficient on inflation is less than 1. He then tests the
model for the period from 1987 - 1997 and finds that the coefficient is greater than 1. He argues that this
result could explain the Great Inflation because higher inflation would cause a decline in the real interest
rate which would increase economic activity and inflation. The switch under Fed Chair Alan Greenspan
meant that the Federal Reserve was raising the real interest rate in response to rising inflation. Clarida,
Galí, and Gertler (1999) estimate a Taylor Rule as well for the pre- and post-Paul Volcker era at the
Federal Reserve (Volcker became chair in 1979). They find the same result as Taylor. However, as they
point out this implies that the New Keynesian model has multiple equilibria. With multiple equilibria,
this leaves open the possibility of self-fulfilling economic fluctuations. In other words, beliefs that have
nothing to do with economic fundamentals can push the economy into another equilibrium.
These estimates, however, have been called into question. In a series of papers, Orphanides (2002,
2003, 2004) calls into question these claims. In particular, he argues that the Taylor rules that were
estimated by Taylor and Clarida, Galí, and Gertler were estimated with ex post data on the output gap.
To accurately estimate the parameters of the rule, consistent with the behavior of policymakers at that
time, we need to use the data that was available to those policy makers. Thus, we need real-time data
and not ex post data. Orphanides finds that when one uses real-time data (from the Fed’s own Greenbook
forecasts), the coefficient on inflation in both the pre- and post-Volcker eras is greater than 1. This implies
that the change in monetary policy between the two eras is not explained by a higher emphasis placed
on inflation, as the previous studies claim. We must look elsewhere for an explanation of the change in
monetary policy.
I have made the argument in my own research that the change in policy was a change in doctrine
(Hendrickson, 2012). During the pre-Volcker era, the Federal Reserve started to think about monetary
policy through the lens of the Phillips Curve. As you might recall from my notes on money earlier in
this course, Fed Chair Arthur Burns states in his diary that controlling inflation requires a combination
of monetary policy to influence activity and wage and price controls. He reiterated this point in public
statements as well. This is clearly not a view consistent with the quantity theory.
I argue that the change when Paul Volcker took over was to place an emphasis on the central bank’s
role in creating inflation.4 In other words, the Taylor rule-view is that during the 1970s the Federal
Reserve did not sufficiently respond to inflation. Think about that statement. This view seems to suggest
that the Federal Reserve is something akin to a firefighter who must arrive on the scene to put out the
fire. However, this then raises the question as to who started the fire. The Federal Reserve is not an
inflation fighter, but rather an inflation creator. Inflation is caused by expansionary monetary policy. The
change under Volcker was a recognition of the Federal Reserve’s causal role in creating inflation. When
Volcker took over, he explicitly took responsibility for inflation and announced that the Federal Reserve
was going to reduce inflation by reducing the growth rate of the money supply.

4 A Note on Getting Money Into a New Keynesian Model

One reason why this model does not include money is that money would not play any meaningful role
in the equilibrium conditions. For example, suppose that we added real money balances to the utility
4
My views on this were heavily influenced by the work of Robert Hetzel at the Richmond Fed and his book on Federal
Reserve policy (Hetzel, 2008).

10
function. By doing so, we could get a log-linearized money demand equation of the form:

`t = α0 yt − α1 Rt

where ` is real money balances, y is income, R is the nominal interest rate and α0 and α1 are parameters.
Note that if we add this to the 3 equation system that we presented above, the inclusion doesn’t add
anything to the implications of the model. In fact, we can solve the 3 equation system and then plug the
solutions for y and R into the money demand equation to get a solution for `.
Some have used this result to argue that money isn’t important. But this is circular logic. Why?
Because we started with the assumption that money is unimportant. We then derived a closed system
of equations. We then added a new equation and a new unknown to check the validity of money. In
doing so, we find that the inclusion of money does not matter to the dynamics of the model. Well, of
course it doesn’t! It is trivial to show that adding one new equation and one new unknown will not affect
the implications of the otherwise closed system. However, this is not because we’ve proven that money is
unimportant, but rather because we initially assumed that money was unimportant.
In order to test whether money is important, you need to construct a model in which money serves a
role consistent with what I discussed in my notes on money. In those notes, we want to have a model in
which money is essential for trade. We could do this by adding a search element to our model. However,
we could also construct a “reduced form” assumption that captures this property. One way to do this is
to assume that money reduces transactions costs, or shopping time. Allow me to explain.
The basic idea when it comes to evaluating money is to capture the key feature of money, which is
that money makes it easier to exchange. How can we get this into our model without changing too much
of the model’s structure. Well, one thing that we could do is assume that households have to shop for
goods. Furthermore, we can assume that the amount of time that the individual spends shopping for
good is a decreasing function of how much money the individual is holding. Thus, we could re-write our
household’s utility function as
∞  1−σ
h1+

t ct
X
t s
E0 β − − ht
1−σ 1+
t=0

where hs is shopping time and

1 Pt ct ω
 
hst =
ω Mt

References
[1] Beckworth, David and Joshua R. Hendrickson. 2017. “Nominal GDP Targeting and the Taylor Rule
on an Even Playing Field.” Working paper.

[2] Calvo, Guillermo. 1983. “Staggered Prices in a Utility-Maximizing Framework.” Journal of Monetary
Economics, Vol. 12, No. 3, p. 383 - 398.

[3] Clarida, Richard, Jordi Galí, and Mark Gertler. 1999. “The Science of Monetary Policy: A New
Keynesian Perspective.” Journal of Economic Literature, Vol. 37, p. 1661 - 1707.

[4] Cochrane, John H. 2011. “Determinacy and Identification with Taylor Rules.” Journal of Political
Economy, Vol. 119, No. 3, p. 565 - 615.

[5] Galí, Jordi. 2008. Monetary Policy, Inflation, and the Business Cycle. Princeton: Princeton University
Press.

11
[6] Hendrickson, Joshua R. 2012. “An Overhaul of Federal Reserve Doctrine: Nominal Income and the
Great Moderation.” Journal of Macroeconomics, Vol. 34, No. 2, p. 304 - 317.

[7] Hetzel, Robert L. 2008. The Monetary Policy of the Federal Reserve: A History. Cambridge: Cambridge
University Press.

[8] Howitt, Peter. 1992. “Interest Rate Control and Nonconvergence to Rational Expectations.” Journal
of Political Economy, Vol. 100.

[9] Howitt, Peter. 2005. “Monetary Policy and the Limitations of Economic Knowledge,” in Colander
(ed.), Post Walrasian Macroeconomics. Cambridge: Cambridge University Press.

[10] Orphanides, Athanasios. 2002. “Monetary Policy Rules and the Great Inflation.” American Economic
Review, Vol. 92, No. 2, p. 115 - 120.

[11] Orphanides, Athanasios. 2003. “The Quest for Prosperity Without Inflation.” Journal of Monetary
Economics, Vol. 50, p. 633 - 663.

[12] Orphanides, Athanasios. 2004. “Monetary Policy Rules, Macroeconomic Stability, and Inflation: A
View from the Trenches.” Journal of Money, Credit and Banking, Vol. 36, No. 2, p. 151 - 175.

[13] Rotemberg, Julio J. 1982. “Monopolistic Price Adjustment and Aggregate Output.” The Review of
Economic Studies, Vol. 49, No. 4, p. 517 - 531.

[14] Rotemberg, Julio J. and Michael Woodford. 1995. “Dynamic General Equilibrium Models with Im-
perfectly Competitive Product Markets,” in Cooley and Prescott (eds.), Frontiers of Business Cycle
Research. Princeton: Princeton University Press.

[15] Taylor, John B. 1999. “A Historical Analysis of Monetary Policy Rules,” in Taylor (ed.), Monetary
Policy Rules. Chicago: Chicago University Press.

[16] Woodford, Michael. 2003. Interest and Prices. Princeton: Princeton University Press.

12