ECON7751 - Macroeconomic

Theory II
Lecture Notes from Susanto Basu’s lectures

Paul Anthony Sarkis

Boston College
Contents

1 Consumption Theory 8

1.1 Introduction: the Fisher model . . . . . . . . . . . . . . . . . . . 8

1.1.1 Preferences of the consumer . . . . . . . . . . . . . . . . 8

1.1.2 Market structure . . . . . . . . . . . . . . . . . . . . . . 9

1.1.3 Budget constraint . . . . . . . . . . . . . . . . . . . . . . 9

1.1.4 Optimality . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.1.5 Intertemporal-substitution and wealth effects . . . . . . . 11

1.2 Consumption under Certainty . . . . . . . . . . . . . . . . . . . 12

1.2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.2.2 Optimality . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.2.3 Explicit solution: polynomial utility case . . . . . . . . . 13

1.3 Consumption under Uncertainty . . . . . . . . . . . . . . . . . . 15

1.3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3.2 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1
 1.3.3 The Permanent Income Hypothesis . . . . . . . . . . . . 22

1.3.4 Assumptions used in the model . . . . . . . . . . . . . . 22

1.3.5 Variable interest rates and CES . . . . . . . . . . . . . . 23

1.4 Empirical evidence and deviations from the PIH . . . . . . . . . . 24

1.4.1 Testing the PIH . . . . . . . . . . . . . . . . . . . . . . . . 24

1.4.2 Liquidity Constraints . . . . . . . . . . . . . . . . . . . . 25

1.4.3 Transaction Costs . . . . . . . . . . . . . . . . . . . . . . 26

1.4.4 Deviations from the PIH . . . . . . . . . . . . . . . . . . 26

1.4.5 Precautionary Saving . . . . . . . . . . . . . . . . . . . . . 27

2 Asset Pricing 28

2.1 Consumption-based Asset Pricing . . . . . . . . . . . . . . . . . 28

2.1.1 From consumption to stock prices . . . . . . . . . . . . 28

2.1.2 More on stock prices . . . . . . . . . . . . . . . . . . . . 30

2.1.3 Consumption beta . . . . . . . . . . . . . . . . . . . . . . 31

2.1.4 Stock prices under different utility functions . . . . . . . . 31

2.2 Testing the C-CAPM . . . . . . . . . . . . . . . . . . . . . . . . 32

2.2.1 Consumption beta or market beta? . . . . . . . . . . . . 32

2.2.2 Equity-premium puzzle . . . . . . . . . . . . . . . . . . 32

2.3 Solutions to the equity-premium puzzle . . . . . . . . . . . . . . . 34

2.3.1 Selection Bias . . . . . . . . . . . . . . . . . . . . . . . . . 34

2
 2.3.2 Habit formation . . . . . . . . . . . . . . . . . . . . . . . . 34

2.3.3 Limited participation . . . . . . . . . . . . . . . . . . . . 35

2.3.4 Disasters and risks . . . . . . . . . . . . . . . . . . . . . 35

2.4 Stock Market Rationality . . . . . . . . . . . . . . . . . . . . . . 35

2.4.1 Arbitrage equation . . . . . . . . . . . . . . . . . . . . . 35

2.4.2 Shiller’s test . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 Real Business Cycle models: an introduction 37

3.1 Stylized facts: what we want from a business-cycle model . . . . . 37

3.2 Setup of the model . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2.1 Housedolds . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2.2 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2.3 Government . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.2.4 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.3 Dynamics and results . . . . . . . . . . . . . . . . . . . . . . . . 43

3.3.1 Shocks to G . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.3.2 Shocks to Z . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.4 Interpretation of the model . . . . . . . . . . . . . . . . . . . . . . 47

3.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.5.1 Fixed labor and capital: partial equilibrium RBC . . . . . 49

3.5.2 Capital shock . . . . . . . . . . . . . . . . . . . . . . . . 49

3
 3.5.3 News shocks . . . . . . . . . . . . . . . . . . . . . . . . 49

4 Advanced RBC models 51

4.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.1.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.1.2 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.1.3 Government shocks . . . . . . . . . . . . . . . . . . . . 53

4.1.4 Technology shocks . . . . . . . . . . . . . . . . . . . . . 53

4.1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.2 Solving the model . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.2.1 Log-linearization . . . . . . . . . . . . . . . . . . . . . . 55

4.2.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.2.3 System of log-linearized equations . . . . . . . . . . . . 59

4.3 Dynamics and results . . . . . . . . . . . . . . . . . . . . . . . . 62

4.3.1 Shocks to Z . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.3.2 Shocks to G . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.4 Interpreting the results . . . . . . . . . . . . . . . . . . . . . . . . 67

4.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

4.5.1 Capital Shocks . . . . . . . . . . . . . . . . . . . . . . . 68

4.5.2 Patience shocks . . . . . . . . . . . . . . . . . . . . . . . 68

4.5.3 Laziness shocks . . . . . . . . . . . . . . . . . . . . . . . 69

4
 4.5.4 News shocks . . . . . . . . . . . . . . . . . . . . . . . . 69

4.5.5 Uncertainty shocks . . . . . . . . . . . . . . . . . . . . . 70

5 Non-Walrasian RBC models 71

5.1 Externalities in production . . . . . . . . . . . . . . . . . . . . . . 71

5.1.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.1.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.1.3 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 78

5.2 Imperfect Competition: the Rotemberg-Woodford model . . . . 79

5.2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

5.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5.2.3 Time-varying markups . . . . . . . . . . . . . . . . . . . 85

6 New Keynesian models 88

6.1 NK model without K, flexible prices . . . . . . . . . . . . . . . . 88

6.1.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6.1.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6.1.3 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 98

6.2 NK model without K, sticky prices . . . . . . . . . . . . . . . . . 99

6.2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

6.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6.3 Three-equations NK model . . . . . . . . . . . . . . . . . . . . . 103

5
 6.3.1 Taylor Rule . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.3.2 Role of the monetary policy . . . . . . . . . . . . . . . . 105

6.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.3.4 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 105

6.3.5 Demand block . . . . . . . . . . . . . . . . . . . . . . . . 105

6.3.6 Supply block . . . . . . . . . . . . . . . . . . . . . . . . 106

6.3.7 Taylor rule: a discussion on the optimal monetary policy . 107

6.3.8 IRFs for NK models . . . . . . . . . . . . . . . . . . . . . 108

6.3.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.4 NK model with investment . . . . . . . . . . . . . . . . . . . . . 113

6.4.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.4.2 IRFs for NKK models . . . . . . . . . . . . . . . . . . . . 116

6
Chapter 1

Consumption Theory

This chapter investigates the theory of consumption, or in other words, how

agents choose the level of consumption based on their preferences subject to
budget constraints. The introduction of the chapter looks at a simple two-period
model and introduces the main topics of consumption theory. The following
sections will look deeper into these concepts by analyzing consumer decisions
under certainty and uncertainty for an infinite horizon.

1.1 Introduction: the Fisher model

1.1.1 Preferences of the consumer

In consumption theory, we assume that any household has preferences defined

over a sequence of consumption levels, for each period in time. To start, we will
consider the case of two-periods (Fisher model). This means that a consumer will
choose two consumption levels (ct and ct+1 ) to satisfy best his preferences.

A household has an instantaneous utility function u(c) representing his pref-

erences at each point in time. We typically assume that u(·) is an increasing
concave function. The lifetime utility of any household will be assumed to be the
net present value of every period’s utility level. In order to discount utility, we

7
assume the household has a subjective rate of time preference, denoted ρ, yielding
1
the discount rate 1+ρ . Already we can see a link between ρ and the ”patience” of
the household. If ρ is high, then the discount factor is low, future utility is not
very important: the household is impatient. When ρ is low, the discount factor is
high: the household is patient.

In the two-period case, we define lifetime utility as:

1
U(ct , ct+1 ) = u(ct ) + u(ct+1 )
1+ρ
This form of utility is called time-separable since any variation of consumption
affects only utility in that period. Note that sometimes we will use β to denote
the discount factor.

1.1.2 Market structure

The consumer faces two markets in this setting, the goods market, in which he
buys consumption goods (c) at a normalized price of 1, and an asset market, in
which he buys assets that yield a return in the next period. The price of the asset
is determined by its return in the next period, denoted r. For now, we assume
that the consumer is a price-taker in both markets, meaning that he does not
internalize the effects of his consumption on prices (the price of consumption
and assets are fixed).

The consumer also gets an income for labor (fixed in this setting), denoted as y.
He gets a wage in both periods.

1.1.3 Budget constraint

In order to solve the model, we have to make sure that the consumer can only
buy with what he has in his budget. Approaching this setting with game theory,
we can be sure that in the last period, the consumer will spend all his assets and
his income in consumption (he has no incentive to save because he will not get
the returns next period). Hence, in the second period, he consumes his income

8
and the value of his assets saved in the first period:

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

In the first period, he can choose to either consume or save his income (assuming
he has no initial wealth). This gives the following budget constraint:

ct + at+1 = yt

We see that both constraints must be satisfied in each period, and that they are
linked by a variable: savings of the first period, at+1 . This fact helps us writing
both constraints as a single, inter-temporal constraint:
1 1
ct + ct+1 = yt + yt+1
1+r 1+r
In words, this constraint also means that the net present value of consumption
must be equal to the net present value of income (also called human wealth).

1.1.4 Optimality

We can now set up the consumer’s problem and solve it:

1 1
max u(ct ) + βu(ct+1 ) s.t. ct + ct+1 = yt + yt+1
ct ,ct+1 1+r 1+r
This problem must satisfy the following two FOCs:

• u0(ct ) − λ = 0
1
• βu0(ct+1 ) − 1+r λ =0
⇒ u0(ct ) = β(1 + r) · u0(ct+1 )

This last equation is called the Euler equation for consumption: it determines the
path of consumption with respect to market parameters. This Euler equation can
also be interpreted in a different way. Consider u0(ct ) as the marginal utility of
consumption in the first period, while β(1 + r) · u0(ct+1 ) is the marginal utility of
consumption in the second period. We can describe the latter also as the marginal
utility of saving in the first period: thus, the optimal path of consumption equates
the marginal utility of consumption and savings.

9
An interesting result of this Euler equation is that the path of consumption is
not determined by income (y does not appear in the Euler equation), this means
that income only defines the level of consumption (through the IBC). To see this
1
clearly, consider the case when ρ = r, then β = 1+r and we have that Euler
equation yields:
u0(ct ) = u0(ct+1 ) ⇔ ct = ct+1
regardless of what happens to income. This result is linked to a few hypotheses
that we will try to let go of in the next sections. In particular, this is the case
because consumers can borrow and lend freely, in a perfect financial market and
also because there is no uncertainty with respect to interest rates, income, etc.

1.1.5 Intertemporal-substitution and wealth effects

In order to investigate comparative static, we will use a specific yet general

1−σ
enough utility function, a CES utility of the form: u(c) = c1−σ which is ln(c)
when σ = 1. Using this functional form, we can interpret σ as the willingness
to smooth consumption: the smaller σ is, the more willing a consumer is to let
their consumption vary over time. We can see that by actually solving the Euler
equation. We know that u0(c) = c−σ , thus:
 σ1
1 +

ct+1 r
u0(ct ) = β(1 + r) · u0(ct+1 ) ⇒ =
ct 1+ρ

Here, the higher σ is, the closer to 1 is the inter-temporal ratio of consumption,
meaning that consumption will be almost equal at all periods.

Using a CES utility function allows us to derive a simple form for the inter-
temporal elasticity of substitution (IES), denoted θ. The IES is defined as:

u0(c) c−σ 1
θ(c) ≡ − = =
00
u (c)c −σc −σ σ
which is a constant for any level of consumption (giving the name CES).

The inter-temporal rate of substitution can give us interesting insight about how
the consumption path reacts to changes in the interest rate. From the Euler
equation, we already know that a higher r implies a steeper consumption path

10
(ct+1 /ct is greater). In fact, it seems logical since 1 + r is in fact the relative price
of ct in terms of ct+1 . Therefore, we could say that in general, a higher r tends to
increase savings.

However, analysis of an increase is not that simple because r also determines the
net present value of wealth and a higher r would also mean lower wealth if the
consumer is borrowing, higher wealth if he is lending, or nothing if he has no
assets. One must consider these cases when looking at the effect of an increase in
r.

Note: a log-utility function (σ = 1) means a one for one substitution of consump-

tion between periods.

1.2 Consumption under Certainty

We have previously introduced the main objective of consumption theory, looking

at consumption decisions with respect to income, interest rates, preferences, etc.,
but we did it in a very simplistic way. The next two sections are going to treat
this topic in a more general way, yet leaving it simple enough. By simple enough,
we mean that we still consider the consumer as a price-taker in markets, his labor
decision is still fixed, and all functions are relatively well-behaved.

1.2.1 Model

A representative consumer seek to maximize lifetime utility by choosing an

optimal path for consumption, subject to instantaneous budget constraints:
∞
Õ
max Ut ≡ β s u(Ct+s ) s.t. Ct+s + At+s+1 = (1 + r)At+s + Yt+s ∀s ≥ 0
{Ct }
s=0

Recall that the utility function used above is additively time-separable.

In this model, At , the asset holdings, is a state variable. We call it this way because
it summarizes all past decisions on consumption, given the past realizations of
income (since not consuming increases A).

11
The income, Yt , is an exogenous variable: it is not linked to any decision and it
fluctuates freely. Even though the same things can be said of β and r, we call
these variables the parameters of the model, since they do not move in time.

Finally, consumption Ct is the choice (or control) variable of the model: the
variable used to solve the model. Note that controlling Ct also means controlling
At+1 because of the instantaneous budget constraint.

1.2.2 Optimality

Since choosing Ct and At+1 is the same thing, we can solve the model by replacing
Ct by its constraint expressed in assets holdings terms and solve it.
∞
Õ
max Ut ≡ β s u((1 + r)At+s + Yt+s − At+s+1 )
{ At+1 }
s=0

which yields the following FOCs, for all s:

−u0(Ct+s ) + β(1 + r)u0(Ct+s+1 ) = 0

which is exactly the Euler equation that we found in the two-period case. Note
that because this equation holds for any s ≥ 0, we can replace u0(Ct+s+1 ) by
any further condition, meaning that at any point in time, the optimal path is
set in stone from the first period. We say that the consumption decision is time
consistent.

1.2.3 Explicit solution: polynomial utility case

In order to look at the consequences of such a model, let’s assume a polynomial

utility function representing the consumer’s preferences:
α1 2
u(Ct ) ≡ α0Ct − C
2 t
Further, we assume that β and r are defined such that β(1 + r) = 1 or equivalently,
we assume that ρ = r.

12
Recall the Euler equation: u0(Ct+s ) = β(1 + r)u0(Ct+s+1 ), now we can rewrite it as:

α0 − α1Ct+s = α0 − α1Ct+s+1

⇔ Ct+s = Ct+s+1
This result implies that the optimal decision for the consumer is to perfectly
smooth consumption: exactly what we found in the previous section with the
CES utility function.

Now this equation, while powerful, does not solve the model. In fact, the solution
to any dynamic model is an expression that links control variables to state and
exogenous variables. In our case, we need to find the expression of Ct in terms of
At , Yt , r and β.

To do that, start with the instantaneous budget constraint:

Ct + At+1 = (1 + r)At + Yt

and write in terms of At :

1 1 1
At = At+1 + Ct − Yt
1+r 1+r 1+r
Since this instantaneous budget constraint holds for any period, we can write:
1 1 1
At+1 = At+2 + Ct+1 − Yt+1
1+r 1+r 1+r
and replace it in the previous budget constraint:

1 1 1 1 1 1
 
At = At+2 + Ct+1 − Yt+1 + Ct − Yt
1+r 1+r 1+r 1+r 1+r 1+r

and we get:
∞
1 1 Õ 1
At = lim AT + (Ct+s − Yt+s )
T→∞ (1 + r)T 1 + r s=0 (1 + r)s

The first term seems scary but if you look at it closely you can see how to simplify
it. Indeed, the term represents the net present value of the consumer’s assets at
the infinite horizon. This value cannot be negative, because of the no-Ponzi-game

13
condition (else, the consumer would roll over debt ad infinitum to consume more
and more). Moreover, the value cannot be positive (again, by a game-theoretic
approach, any rational agent would consume all his wealth in the last period).
Then, it must be that the term is equal to 0. We are then left with two terms: the
net present value of consumption, and the net present value of income (human
wealth). We can write:
∞
1 Õ 1
At + Ht = Ct+s
1 + r s=0 (1 + r)s

Using the Euler equation of earlier, we know that consumption is perfectly smooth,
meaning that we can write Ct for any Ct+s . This allow us to simplify or equation
to:
1 1
At + Ht = Ct 1
1 + r 1 − 1+r
Ct
⇔ At + Ht =
r
⇔ r(At + Ht ) = Ct
In the same way, if we assume that Yt is constant for all periods, we can write the
exact definition of consumption as:
Ct = r At + Yt
which replaced in the budget constraint gives At = At+1 : wealth is constant over
time.

1.3 Consumption under Uncertainty

1.3.1 Model

Again, we use a representative household in an economy such that he can borrow

and lend freely (no transaction costs) at a fixed interest rate r. The horizon is
infinite (Ramsey model) so as before we get:
"∞ #
Õ
max Et β s U(Ct+s ) s.t. Ct+s + At+s+1 = (1 + r)At+s + Yt+s
Ct
s=0

14
However, this time we assume that consumers do not know future realizations
of variables. In particular, we say that the variable Yt (the only variable that con-
sumers have no control on) is stochastic. Agents know the probability distribution
of Yt (its stochastic process). Therefore, in order to solve the problem, we assume
that consumers have rational expectations.

Rational Expectations

Agents formulate expectations in such a way that their subjective probability

distribution of economic variables (conditional on the available information) co-
incides with the objective probability distribution of the same variable (according
to a measure of the state of nature) in an equilibrium. This means that agents
have perfect foresight, under uncertainty.

As we’ve seen above, we represent these expectations by using the notation

Et [·] ≡ E [·|It ], the expectation of any variable, conditional on the information
set at time t.

Rational expectations have some interesting properties:

• No systematic bias: Et [Ct+1 − Et [Ct+1 ]] = Et [εt+1 ] = 0

• No autocorrelated bias: Cov (εt+s, εt ) = 0 for all s , 0

• Law of iterated expectations: Et [Et+s [Xt+v ]] = Et [Xt+v ] for all v ≥ s ≥ 0

Optimal path à la Basu

This time we use the method of perturbation (Basu’s favorite) in order to find the
optimal path. Let δ > 0 be a variation so small in level that U 0(x + δ) ≈ (δ)U 0(x).
Now, assuming you are on the optimal path, consider δ to be an amount that you
choose not to consume today in order to consume it tomorrow. That is, you save
δ more today and get to consume δ(1 + r) more on next period, on in other words,
you take off from the optimal path. Because this perturbation takes you out of the

15
optimal path, the reallocation of δ has to incur a loss in utility. Therefore we get:

1+r 0
 
(−δ)U (Ct ) + Et (δ)
0
U (Ct+1 ) ≤ 0
1+ρ

In the same manner, a positive perturbation of δ will also cause a loss in utility,
hence:
1+r 0
 
(δ)U (Ct ) + Et (−δ)
0
U (Ct+1 ) ≤ 0
1+ρ
You can clearly see that both equations are perfect oppposites of each other,
implying that:
1+r 0
 
−U (Ct ) + Et
0
U (Ct+1 ) = 0
1+ρ
This is the Euler equation for consumption, under uncertainty.

Consumption as a random walk

In order to let the model give us more insight on consumer behavior, we will have
to make some assumptions. First, we impose a restriction that rt = r, meaning
that the interest rate is fixed. This will of course be true is the steady-state.
Second, we also assume that this interest rate is equal to ρ, or equivalently, that
β(1 + r) = 1, which will also be satisfied in equilibrium. These two assumptions
yield a simplified version of the Euler equation:

−U 0(Ct ) + Et [U 0(Ct+1 )] = 0

⇔ U 0(Ct ) = Et [U 0(Ct+1 )]
This only gives one (not very informative) piece of information: the optimizing
consumer will choose a path of consumption so that his expected marginal utility
is the same for all period: he smooths consumption over time.

In order to simplify further the previous Euler equation, Hall (1978) assumes that
utility follows a quadratic form with a satiation point at C̄. If C̄ is far enough
from the typical level of consumption, we have that U(·) is concave.

16
In particular, let U(Ct ) = −(C̄ − Ct )2 . Then,
U 0(Ct ) = Et [U 0(Ct+1 )]
2(C̄ − Ct ) = Et 2(C̄ − Ct+1 )
 

Ct = Et [Ct+1 ]
Ct+1 = Ct + εt+1 where Et [εt+1 ] = 0
This result is one of the most important in consumption theory. Note that it
resembles closely the two previous results implying perfect smoothness of con-
sumption, this time adding the fact that income is uncertain. This result has three
main implications:

1. Current consumption alone is the best predictor of future consumption.

This comes from the fact that the error term is i.i.d. with mean zero.
2. Changes in consumption are ex-ante unpredictable. In fact, if you look at
Et [Ct+1 − Ct ], you get:
Et [Ct+1 ] − Ct = 0
3. Consumption follows a random walk. This is the basic result that comes
from the form of the equation.

These consequences will be tested in a follow-up discussion on the permanent

income hypothesis which includes all these results in an interesting and useful
narrative.

Explicit solution

Now that we know what variations in the consumption will look like (we know
that consumers will smooth), we are interested in the actual consumption function.

Starting from the budget constraint, we have that:

At+1 = (1 + r)At + Yt − Ct
where all variables are known at time t (remember that At+1 is the decision made
at time t); we rewrite this equation as a forward looking equation as:
1 1 1
At = At+1 + Ct − Yt
1+r 1+r 1+r

17
which must hold at all time, in particular at time t + 1:
1 1 1
At+1 = At+2 + Ct+1 − Yt+1
1+r 1+r 1+r
Now we know At+1 , but we do not know anything about the other variables. We
can nevertheless use rational expectations to clear things out:
1 1 1
At+1 = Et [At+2 ] + Et [Ct+1 ] − Et [Yt+1 ]
1+r 1+r 1+r
Replacing this last equation in the contemporaneous budget constraint yields:

1 1 1 1 1 1
 
At = Et [At+2 ] + Et [Ct+1 ] − Et [Yt+1 ] + Ct − Yt
1+r 1+r 1+r 1+r 1+r 1+r
1 1 1 1 1
⇔ At = 2
Et [At+2 ]+ 2
Et [Ct+1 ]− 2
Et [Yt+1 ]+ Ct − Yt
(1 + r) (1 + r) (1 + r) 1+r 1+r
This process can be repeated many times, until the limit:
T
1 Õ 1
At = lim Et [AT ] + E [Ct+s − Yt+s ]
s+1 t
T→∞ (1 + r)T
s=0
(1 + r)

Using the no-Ponzi game condition, we can restrict the value of assets to be equal
to zero at the limit. Adding the expected NPV of income of both sides, we get
that the expected NPV of consumption is equal to the assets holdings at time t
plus the expected NPV of labor (or human wealth):
" T # " T #
1 Õ 1 1 Õ 1
Et C
s t+s
= At + Et Y
s t+s
1+r s=0
(1 + r) 1 + r s=0
(1 + r)

Now, because Ct = Et [Ct+1 ] = Et [Ct+s ] for all s, we have that:

" T # T
1 Õ 1 1 Õ 1
Et Ct+s = Ct
1+r s=0
(1 + r) s 1 + r s=0 (1 + r)s

1 1 1 1+r
= Ct · 1
= Ct ·
1 + r 1 − 1+r 1+r r

18
We can replace this last bit in the previous equation to get:
" T #
1 1+r 1 Õ 1
Ct · = At + Et Y
s t+s
1+r r 1+r s=0
(1 + r)
" T #
r Õ 1
Ct = r · At + Et Yt+s
1+r s=0
(1 + r)s

1.3.2 Dynamics

Shocks to the interest rates

We have seen that the consumption path is equal to the current value of assets plus
the expected net present value of labor income (human wealth). This equation
has a lot going on for it, therefore making it tricky to understand intuitively the
effects of certain shocks to the economy. In the particular case of interest rates,
we have multiple effects showing up:

• Wealth effects: an increase in the interest rate will reduce the expected net
present value of income, regardless of its level. Therefore, this has an effect
of reducing consumption (income is lower). ⇒ C ↓

• Substitution effects: an increase in the interest rate will also increase po-
tential returns on assets, making savings relatively more interesting than
consumption. This will have an effect of transferring money from consump-
tion to assets.⇒ C ↓

• Income effects: an increase in interest will also have effects on income. In

fact, a higher interest rate yields a higher return on assets, in turn making
more consumption affordable.⇒ C ↑

In the particular case of log-utility, the last two effects are known to exactly offset
each other, leaving only a negative wealth effect on consumption, following an
increase in interest rates.

19
Shocks to preferences

Shocks to income

Recall that we have found that consumption is determined by the following

equation: " T #
r Õ 1
Ct = r · At + Et Yt+s
1+r s=0
(1 + r)s

Now, assume a stochastic process for income Yt such that Yt has a fixed component
Ȳ and a stochastic component Ỹt given by Ỹt = ρỸt−1 + εt . The process just
mentioned allows us to compute several expectations of income:

• Et [Yt ] = Yt = Ȳ + Ỹt

• Et [Yt+1 ] = Ȳ + Et Ỹt+1 = Ȳ + ρỸt

 

• …

• Et [Yt+s ] = Ȳ + ρsỸt
hÍ i
1
Therefore, we can write r
1+r Et T
s=0 (1+r)s Yt+s as:
" T #
r Õ 1 r 1 r
Et [Ȳ + ρ s
Ỹt ] = Ȳ + · ρ Ỹt = Ȳ + Ỹt
1+r s=0
(1 + r) s 1 + r 1 − 1+r
1 + r − ρ

if and only if ρ < 1 + r which should be the case.

This specification allows for a simple representation of shocks on income. For

example, let there be a pure transitory shock on income, such that ρ = 0, then
∂Ct
∂Yt = 1+r , which is a very small amount. This phenomenon is due to the fact that
r

consumers want to smooth consumption over time, disconnect their consumption

from small transitory shocks. Now, if the shock is permanent, as in ρ = 1, then
∂Ct
∂Yt = 1, meaning that the response to that shock is one-to-one.

20
1.3.3 The Permanent Income Hypothesis

The last equation that was derived is the fundamental equation of the Permanent
Income Hypothesis. We have seen that this equation implies small MPC following
transitory shocks but bigger values for permanent shocks on income. This would
mean, for example, that tax returns should not yield a high increase in consump-
tion, since it is a transitory shock in income, that is also small in comparison to
lifetime income. We would therefore expect small multipliers in these cases. This
also means that a government that wants to boost consumption via tax returns
needs to commit to a large return that would stick in time.

1.3.4 Assumptions used in the model

Hall’s model as described previously relies on a list of assumptions that help

with solving the model. However, to see if the model is perfectible, we can try to
relax some of the assumptions. First, let’s allow ourselves to review them in depth.

(1). Consumers optimize : of course, this assumption is typically the basis

for economic research because it allows for easy formalization. Nevertheless,
one can think about the true nature of this assumption as in : are people really
thinking about this stuff? Is aggregation enough to justify an optimizing behavior,
on average?

(2). Consumers have rational expectations : again, this assumption presents

the advantage of facilitating the formalization but its presence in individual be-
havior can be discussed.

(3). Consumers can borrow and lend, freely, at the same rate rt : this as-
sumes that every consumer, regardless of their income, their wealth, their previous
record or their use of money (borrowed or lent) will get the same rate from the
bank. It is obvious that it is not the case in real life where bank lend you money
for close to 20% while borrows from you at a 1,50% rate.

(4.) All consumption is non-durable :

(5.) Consumers face no transaction cost : in the same line of argument than

21
(3.), this assumption relies on the fact that consumers can behave freely, without
having to pay for anything (banking fees, taxes, etc.)

(6.) Utility can be measured as a function of form U(C) (or U(C) + V(X)) :
this is important because it states that nothing else than consumption can affect
utility in an interactive manner.

(7.) Utility is additively time-separable : this assumption means that future

or past consumption cannot affect our utility function in a way that cannot be
separated from current consumption.

(8.) Discounting occurs at a constant, exponential rate : assumes the form

our discounting function

(9.) rt = r = ρ : this ass. allows for easier computations and should hold at
the steady state under certain conditions, it is not a general enough assumption.

(10.) Utility is quadratic : again, for easier computations (not having to work
around Jensen’s inequality.

1.3.5 Variable interest rates and CES

This first elements that change in our analysis will be variable rt and a CES
utility function. A variable interest rate will cancel the simplification of the Euler
equation in the same manner as before. This is because rt , ρ. From our basic
FOC:
1 + rt U 0(Ct+1 )
 
Et =1
1 + ρ U 0(Ct )
Now, assuming a CRRA utility function with CES of σ, our FOC becomes :

1 + rt  Ct+1  −σ
 
Et =1
1 + ρ Ct

22
Simplifying, we have:

1 + rt  Ct+1  −σ
 
Et =1
1 + ρ Ct
1 + rt  Ct+1  −σ
= 1 + εt+1
1 + ρ Ct
rt − ρ − σ(ln(Ct+1 ) − ln(Ct )) ≈ εt+1
1
ln(Ct+1 ) − ln(Ct ) ≈ (rt − ρ) + εt+1
σ
We get a final form for the variation of consumption which is very similar to
our result in the Ramsey model. Indeed, it is its approximation in discrete time.
This particular equation is relaxed of two assumptions (variable interest rate and
quadratic utility function) and still gives us a similar Euler equation as before.

1.4 Empirical evidence and deviations from the

PIH

1.4.1 Testing the PIH

We have seen that the Euler equation above gives us, for any CRRA utility function:
1
ln(Ct+1 ) − ln(Ct ) ≈ (rt − ρ) + εt+1
σ
Any simple regression (even OLS) should give you significant results in this form.
Nevertheless, consider the case where interest rates, while variable, are not known
at period t but rather at t + 1. Then the equation above can be written as:
1
ln(Ct+1 ) − ln(Ct ) ≈ (rt+1 − ρ) + εt+1
σ
OLS is not consistent anymore if somehow ε and r are correlated. To counter this
issue, we could use an instrument (as always). In particular, using past values of
the interest rate rt , rt−1, ... should work as they hold no correlation with εt+1 by
definition.

23
Back to the case with known interest rates, we should see that when controlling
for rt and ρ, no information available at time t should affect consumption growth.
This is also true for expectations of future income since they are formed at time
t. However, this should strike you as weird since in everyday life it would be
ridiculous to not base your future consumption out of your expectations of income
(think of buying a house, etc.). This issue leads consumption theory research since
Hall (1978).

There have been a huge number of tests of this hypothesis using microdata, to
name a few of them:

• Carroll & Summers (1991): they show that permanent income hypothesis
is inconsistent with the grossest features of cross-country and cross-section
data on consumption and income.

• Souleles (1999): he finds significant evidence of excess sensitivity in the re-

sponse of households’ consumption to their income tax refunds (predictable
income).

• Johnson-Parker-Souleles (2006): they find that, following the Bush tax

rebates of 2001, households spent 20 to 40 percent of their rebates on
nondurable goods during the three-month period in which their rebates
arrived, and roughly two-thirds of their rebates cumulatively during this
period and the subsequent three-month period.

These authors and others interpret these inconsistencies as effect of two main
factors : liquidity constraints and/or transaction costs.

1.4.2 Liquidity Constraints

Consumers are subject to liquidity constraints, meaning that it can be difficult for
them to borrow when they expect their income to grow (as the PIH would require).
This is a consequence of banks being reluctant to risk and borrowing rates being
higher than lending rates. This leads to a correlation between consumption and
income when income is rising, implying a high marginal propensity to consume.

Gourinchas-Parker (2002) show that consumers seem to be indeed constrained

financially until the age of 40, when they start developing a PIH-type of behavior.

24
This would lead to this type of consumption pattern:

1.4.3 Transaction Costs

Another cause of invalidating the PIH in certain contexts would be the existence
of transaction costs. The argument basically says that, being purely rational,
consumers would apply the Envelope Theorem. This says that an infinitesimal
change in outcome will affect utility the same no matter where you spend it. For
Cochrane (1989), what’s 600$ over a lifetime of income? An infinitesimal change.
This would explain why people would not seek to optimize when receiving small
amounts.

This would imply that data could confirm the PIH for larger, predictable amounts
of money. Hsieh (2003) finds that households in Alaska do not overreact to
payments from the Permanent Fund, which are quite important; whereas he finds
find that the consumption of the very same households is excessively sensitive to
their income tax refunds, as explained before.

Then comes the question : what matters the most for the PIH, income shocks
being large or predictable?

1.4.4 Deviations from the PIH

If we know that the PIH is not very efficient for finding the MPC out of income
shocks, at least for small amounts, we should develop strategies to account for
these deviations.

Campbell and Mankiw (1989) use aggregate data on consumption to model devia-
tions from the PIH. Using the same consumption function as is Hall (1978), they
create two groups of consumers :

• The rule-of-thumb consumers: for whom C1,t = Y1,t where Y represents

disposable income. They constitute a fraction λ of total population.

• The PIH consumers: for whom C2,t − C2,t−1 = εt .

25
Aggregating both behaviors, we get

∆Ct = ∆C1,t + ∆C2,t

∆Ct = λ∆Y1,t + εt

This final equation can be estimated with OLS.

Basu and Kimball (2002) show that, adding labor Nt into the utility function gives
the following Euler equation:

∆ ln(Ct ) = s(rt − ρ) + (1 − s)τ∆ ln(Nt ) + εt

1.4.5 Precautionary Saving

For now, uncertainty has been out of the question because it doesn’t affect any
quadratic utility form à la Hall (1978). However, considering other utility functions
might lead to further implications of uncertainty.

Assume a general utility function U(Ct ) where r = ρ:

26
Chapter 2

Asset Pricing

2.1 Consumption-based Asset Pricing

2.1.1 From consumption to stock prices

Recall from the previous chapter that the typical Euler equation looks like this:
1 + rt+1 U 0(Ct+1 )
 
Et =1
1 + ρ U 0(Ct )
where 1 + rt+1 is the return on an asset used for savings. In the particular case of
this chapter, we will assume that the vehicle of savings is a risky asset designed
as a stock.

To hold such an asset, you pay a price pt , which gives you the right to earn a
dividend dt+1 next period, as well as selling it at its new price pt+1 . Hence the
following equation for the total return:
pt+1 + dt+1 pt+1 dt+1
1 + rt+1 = = +
pt pt pt
|{z} |{z}
Capital Gains Dividend-price ratio

Note that because the price is not known perfectly, 1 + rt+1 is actually unknown
at time t. Before coming back to the Euler equation, we will define a term that

27
we encounter often, in order to simplify computations: the stochastic discount
U 0 (Ct+1 )
factor Mt+1 ≡ (1+ρ)U 0 (C ) ; again, this term is also unknown at time t.
t

Then, if we replace the rate of return as well as the stochastic discount factor we
get:
1 = Et [(1 + rt+1 ) · Mt+1 ]
We know from the statistics class that for any two random variables X and
Y , E [XY ] = E [X] E [Y ] + Cov (X,Y ). This rule can be used with the rational
expectations as well so that:

1 = Et [(1 + rt+1 ) · Mt+1 ] = Et [(1 + rt+1 )] · Et [Mt+1 ] + Covt ((1 + rt+1 ), Mt+1 )

And if we multiply by pt :

pt = Et [(pt+1 + dt+1 )] · Et [Mt+1 ] + Covt ((pt+1 + dt+1 ), Mt+1 )

which means that the value pt of any risky asset is determined by:

• The expected price and dividend of the asset, discounted by the stochastic
discount factor.

• The covariance between the returns and the stochastic discount factor
(marginal utility of consumption).

This result is known as the Consumption-based Capital Asset Pricing model, or

C-CAPM.

This theory can be quite useful to link the consumption model as we’ve seen it in
the previous chapter with the setting of financial markets. For example, suppose
two stocks A and B are available : stock A pays off when the economy is in boom
(when consumption is high) while stock B pays a constant rate (regardless of
consumption). Which one should be more valuable?

If we assume that both stock yield the same stochastically discounted return,
then stock B will actually be valued more, because its return is more closely
correlated to the SDF. In fact since it pays off the most during recessions (when
marginal utility is the highest = consumption is the lowest), the second term of
our C-CAPM model will be higher for stock B. In the same direction of ideas, note
that any risk-free asset (delivering a return without correlation to any other state)

28
would have a 0 covariance with the SDF, yielding the following price equation:
1
1 = Et (1 + r f ,t+1 ) · Mt+1 ⇔ 1 + r f ,t+1 =
 
Et [Mt+1 ]

2.1.2 More on stock prices

The previous definition of the C-CAPM was a good introduction to the link
between consumer behavior and stock markets, but it lacks a financial view. The
following subsection will try and look at the problem with more finance in mind.

Going back to the Euler equation, let’s try and solve it recursively:

1 + rt+1 U 0(Ct+1 )
 
1 = Et
1 + ρ U 0(Ct )
pt+1 + dt+1 U 0(Ct+1 )
 
1 = Et
pt (1 + ρ)U 0(Ct )
(pt+1 + dt+1 )U 0(Ct+1 )
 
pt = Et
(1 + ρ)U 0(Ct )
U 0(Ct+1 ) U 0(Ct+1 )
 
pt = Et dt+1 + pt+1
(1 + ρ)U 0(Ct ) (1 + ρ)U 0(Ct )
U 0(Ct+1 ) U 0(Ct+2 ) U 0(Ct+2 )
 
pt = Et dt+1 + dt+2 + pt+2
(1 + ρ)U 0(Ct ) (1 + ρ)U 0(Ct ) (1 + ρ)U 0(Ct )
..
.
" T 
!#
U 0(Ct+s ) U 0(Ct+T )
Õ 
pt = Et lim · dt+s + · pt+T
T→∞
s=1
(1 + ρ)U 0(Ct ) (1 + ρ)U 0(Ct )

This equation tells us that the price of any risky asset is given by two elements:

• The expected ”perceived” net present value of dividends: this is the expected
net present utility of all future cash flow earned by holding the asset. It
can be viewed as a sort of ”personal” net present value of cash flow. This
element is the fundamental part of the price (because you actually get the
money at some point).

29
 • The expected limiting value of the stock: this represents the value of the
stock at the infinite horizon. For simplicity, and to avoid any bubble for-
mation, we assume this term to be zero. This term is known as the bubble
term (you never get the money from that since it is the price at the infinite
horizon).

2.1.3 Consumption beta

Now, combining two results from the previous subsections, the classical C-CAPM
formula and its form for risk-free assets, we can draw even closer connections
between the C-CAPM and the traditional CAPM.

We had:
1 = Et [1 + rt+1 ] · Et [Mt+1 ] + Covt ((1 + rt+1 ), Mt+1 )
1
1 + r f ,t+1 =
Et [Mt+1 ]
which together can give:

Et [1 + rt+1 ]
1= + Covt ((1 + rt+1 ), Mt+1 )
1 + r f ,t+1

⇔ −(1 + r f ,t+1 ) · Covt ((1 + rt+1 ), Mt+1 ) = Et [1 + rt+1 ] − (1 + r f ,t+1 )

⇔ Et [rt+1 ] − r f ,t+1 = −(1 + r f ,t+1 ) · Covt ((1 + rt+1 ), Mt+1 )
| {z }
risk-premium

which looks a lot like the market-beta formula from the CAPM.

2.1.4 Stock prices under different utility functions

Let a representative consumer have the following log-utility function:

30
2.2 Testing the C-CAPM

2.2.1 Consumption beta or market beta?

2.2.2 Equity-premium puzzle

Mehra and Prescott (1985) was the first paper to evaluate the C-CAPM in compar-
ison to its ability to replicate basic data. While it showed that the C-CAPM had a
lot of difficulties to match the data, it opened the way to an immense literature
that goes on again today. Their main finding is the so-called equity-premium
puzzle.

To see it, let a representative consumer have the following CRRA utility function:
C 1−σ −1
u(Ct ) = t1−σ . Then, look back at the Euler equation for a particular stock i:
" #
1 + rt+1
i
U 0(Ct+1 ) −σ 

Ct+1
Et = 1 ⇔ E (1 + rt+1 ) −σ = 1 + ρ
i
1 + ρ U 0(Ct ) Ct

⇔ E (1 + rt+1
i
)(1 + gC )−σ = 1 + ρ
 

Let’s now take the second order Taylor expansion of the term inside the expecta-
tion, around (r i, gc ) = (0, 0).
1
(1 + r i )(1 + gc )−σ ≈ 1 + (r i ) − σgc + −2σ(r i )(gc ) + σ(1 + σ)gc2

2
σ(1 + σ) 2
≈ 1 + (r i ) − σgc − σ(r i )(gc ) + gc
2
Plugging it back into the expectation operator, we get
  1
E r −σ E [g ]−σ[E r E [g ]+Cov r , gc ]+ σ(1+σ)[E [g c ]2 +Var [g c ]] ≈ ρ
 i c
 i c i
2
And since E [g] E r and E [g]2 are so small, we drop them and we get:
 i

  1
E r ≈ ρ + σ E [g ] + σ Cov r , gc − σ(1 + σ) Var [g c ]
 i c i
2

31
This formula is very important because, in order to analyze the difference between
two assets i and j, it gives the equation
 
E r i − E r j = σ Cov r i − r j , gc
   

which is easily testable in an econometric model.

For example, let i denote average stock returns while j would the average return
on government bonds. Then, between 1890 and 1999, the difference in mean return
would be 0.06 while the covariance between this difference and consumption
growth would be 0.002. This implies a σ of 30. Is it plausible?

⇒ This is the equity premium puzzle! (See Mehra-Prescott, 1985).

Signification of σ
Consider two lotteries:

• The first lottery pays $50,000 with p = 0.5 and $100,000 with p = 0.5.

• The second lottery pays $X with certainty.

What values of X makes you indifferent between the two lotteries?

• σ = 0 ⇒ X = 75000

• σ = 2 ⇒ X = 70000

• σ = 3 ⇒ X = 65000

• σ = 30 ⇒ X = 51200

This implies that it is highly unlikely that an individual would have σ = 30,
therefore shadowing doubts on the equity premium equation.

The previous theory also implies a risk-free-rate puzzle (Weil, 1989):

1
ln(Ct+1 ) − ln(Ct ) = (r f − ρ) + et+1 ⇒ E r f = ρ + σ E [gc ]
 
σ
The problem is that, beginning from the post-war period, the average risk-free
rate has been approx. 1 percent, while the consumption growth rate has been
close to 1.5 percent. This cannot be consistent with values of 30 for σ.

32
2.3 Solutions to the equity-premium puzzle

Since the puzzle about asset returns exists in the observed behavior of the indi-
viduals, it does not rely on features of GE other than consumption. From this
argument, there are three approaches to try and explain the puzzle:

• The true difference in mean returns is smaller

• The effective σ is indeed large

• The true Cov () is larger than it looks

2.3.1 Selection Bias

First, one might ask why would the difference in return seem higher than it is?
Brown, Goetzman and Ross (1995) argue that there is a selection bias: we use
data only from markets which have survived for a long time. This is called the
survivor bias and might imply that the stock returns considered are indeed way
higher than the true population returns.

2.3.2 Habit formation

Second, we might argue that the effective σ could be larger. In fact, if we consider
habit formation such that:
(Ct − αCt−1 )1−σ
U(·) =
1−σ
where α is close to 1. Then, we would have very risk-averse behavior, even when
σ = 3. However, this argument does not explain why people would be so risk-
averse, it just finds a way to show that with certain restrictions, aversion to risk
can be found with low values of σ.

33
2.3.3 Limited participation

Finally, the Cov () might not be computed on good values. Indeed, not everyone
holds financial assets, therefore it is a stretch to consider aggregate consumption
instead of consumption from people who actually hold stocks (Mankiw and Zeldes,
1991). Spending on luxury goods for example has a very high correlation with
stock returns (Parker, Ait-Sahlia & Yogo, 2004).

2.3.4 Disasters and risks

2.4 Stock Market Rationality

2.4.1 Arbitrage equation

The question we should ask ourselves here is whether this theory of asset pricing
is consistent with investment decisions made by firms.

Assuming for now that r is constant, we have:

Et [pt+1 + dt+1 ]
pt =
1+r
By recursion, we can get
Et [pt+1 + dt+1 ] Et [dt+1 ] Et [pt+1 ]
pt = = +
1+r 1+r 1+ r
Et [dt+1 ] pt+2 + dt+2

= + Et
1+r (1 + r)2
Et [dt+1 ] Et [pt+T ]
 
dt+2
= + Et 2
+ . . . + lim
1+r (1 + r) T→∞ (1 + r)T
"∞ #
Õ dt+s Et [pt+T ]
= Et + lim
s=0
(1 + r)s T→∞ (1 + r)T

where the first part of the equation is the fundamentals (discounted future cash
flows) and the second part is the limit of the selling price. In Ramsey-type models,
the transversality condition implies that the limit of the stock price must be 0.

34
⇒ This assumption gives us the Efficient Market Hypothesis: the price of a stock
is determined only by the fundamentals.

2.4.2 Shiller’s test

We can follow Shiller (1981) to test this EMH. The basic argument is that if markets
are fully efficient, then the ex-post optimal price based on fundamentals pt∗ is
equal to the expectation of the price Et [pt ]. Therefore,

pt∗ + et+1 = pt ⇒ pt∗ = pt + ut+1

where Cov (pt , ut ) = 0. This implies in turn that Var pt∗ ≥ Var [pt ].
 

Shiller’s test (1981)

Shiller constructed a theoretical p∗ based on ex-post sum of dividends. In
order to include dividends not paid yet (recall that the dividends go to
infinity), he uses an estimate of the trend. The variance bound implied by
the null hypothesis is strongly rejected, even though some economists think
it might be less severe than it seems.

35
Chapter 3

Real Business Cycle models: an

introduction

3.1 Stylized facts: what we want from a business-

cycle model

We have seen last semester a whole class of real (as in not nominal) neo-classical
models of growth. The issue with those models is that they cannot introduce
any short-term kind of analysis. In order to do that, we need to introduce some
shocks in the economy. Real Business Cycles (RBC) models do exactly that: by the
means of real shocks (we will see later how nominal shocks are ineffective), we
can understand how an otherwise stable economy (steady-state) can reproduce
short-term growth or recessions following some shocks.

In the reality of our economy, it seems that after detrending, all variables (con-
sumption, investment, hours worked, wages, etc.) are pro-cyclical, meaning they
move together with output. This means that in essence, we would like a model
that replicates these comovements in the variables, following a shock. We also
observe that investment is more volatile than output, which is more volatile than
consumption. σI > σY > σC . Hours worked is as volatile as output, and finally,
productivity (measured as the Solow residual or labor productivity) is strongly
correlated with output.

36
3.2 Setup of the model

3.2.1 Housedolds

Assumptions

In this model, consumers will be summarized as a representative household

deciding to consume/save and to work.

• The three decision variables are Ct , Bt+1 and Ht (hours worked)

Ct1−σ
• Utility from consumption is a CRRA function: 1−σ

• Utility from leisure is V(H̄ − Ht )

• Investment (Savings) is only available in consumption bonds Bt yielding

a rate rt ; however we will see that the market for these bonds will not be
active as there is only one household (no one to exchange the bonds with).

• The earnings on labor are given by Wt which is set in the labor market.

• The ousehold receives a ”profit” Xt or dividend from firm ownership (not

important right now) and pays a lump-sum tax equal to Tt .

Solving the problem

Hence, the problem of the consumer is written as:

Õ  C 1−σ
"∞ #
t+s
max Et βs + V(H̄ − Ht )
Ct ,Ht ,Bt+1
s=0
1 − σ

s.t. Bt+1 = (1 + rt )Bt + Wt Ht + Xt − Ct − Tt

This yields three FOCs:
Ct−σ = λt (3.1)
V 0(H̄ − Ht ) = Wt λt (3.2)

37
 λt = Et [β(1 + rt+1 )λt+1 ] (3.3)

The first two equations describe the optimal decisions of consumption and leisure,
as they equal the marginal utility from the two, to their respective shadow prices.
The third equation is the now-known Euler equation, which is clearer written as:
   −σ 
Ct+1
1 = Et β(1 + rt+1 )
Ct

3.2.2 Firms

Assumptions

This model considers firms as a separate agent, behaving under multiple assump-
tions:

• Firms are in a perfect competition context where they are price-takers (they
don’t have a profit).

• The production function is the standard neoclassical CRS function: Zt F(K̄, Ht ).

• Physical capital stock is fixed (since consumer cannot invest in it and firms
run no profits).

• Profits are discounted by the stochastic discount factor since consumers

own the firms.

• Firm will choose labor to maximize the expected net present value of their
cash flows.

Solving the problem

The complete problem of the firm is therefore:

"∞ #
Õ
max Et


S(s) Zt+s F(K̄, Ht+s ) − Wt+s Ht+s
Lt
s=0

38
Since firms do not hold any intertemporal assets, we can solve for the static
version of the same problem:

max Zt F(K̄, Ht ) − Wt Ht
Ht

which gives the following FOC:

Zt F 0(K̄, Ht ) = Wt (3.4)

Firms do have a capital stock, even though it is not variable, which earns its
competitive return Rt equal to its marginal productivity (which is variable):

Rt = Zt FK (K̄, Ht )

Since households own the capital stock of the firm and get the return on this
capital: Xt = Rt K̄.

Technology

Finally, we assume that technological progress (not the level but the growth rate)
Z̃t is a stochastic process subject to either temporary or permanent shocks. In
particular, we write:

Z̃t = ρ Z̃t−1 + εtZ where E εtZ = 0

 

To model a temporary shock, we use ρ < 1 or even ρ = 0 for one-time shocks.

Permanent shocks have ρ = 1.

3.2.3 Government

Here, the role of government is pretty straightforward, it taxes the exact amount
it spends every period,
Tt = Ḡt
Expenses Gt are also subject to shocks that are either permanent or transitory.

39
3.2.4 Equilibrium

Equilibrium Conditions

Let’s review all equilibrium conditions and describe how they will interact with
each other:

• Consumption demand: Ct−σ = λt

• Labor supply: V 0(H̄ − Ht ) = Wt λt
h   −σ i
• Euler equation: 1 = Et β(1 + rt+1 ) CCt+1
t

• Aggregate constraint: Yt = Ct + Gt
• Labor demand: Zt F 0(K̄, Ht ) = Wt
• Production function: Yt = Zt F(K̄, Ht )

Solving the problem

Now, we have theoretically three markets: the goods market, the capital market
and the labor market. However, since the capital stock is fixed (i.e. there is no
vehicle of savings in this economy), we can actually reduce the dimensions to
two markets. If you recall general equilibrium classes, with two markets, only
one has to clear in order for the whole economy to clear. This is why we reduce
our model to a single ”consumption-labor” market. This simple version of the
RBC is still in the partial equilibrium literature.

To see that there is actually no assets, consider the fact that capital stock is fixed,
and while consumption bonds do exist, there is only one household, hence no
exchange. We have that S = I = 0 and B = 0.

This means we need to write both the borrowing and labor markets in one general
economy market.

First, from the optimality conditions, we can rewrite:

V 0(H̄ − Ht ) = Wt · Ct−σ

40
which links consumption’s optimal decision to the labor optimal decision. More-
over, we can replace the wage Wt by its expression in the firm’s FOC: Zt F 0(K̄, Ht ).
This yields the final equation:

Zt F 0(K̄, Ht )
Ct−σ = 0
V (H̄ − Ht )
This equation is called the optimality (OPT) condition (as it is derived from both
problems’ FOCs). Note that written in this way, you can see that as hours worked
increase, leisure decreases and marginal utility of leisure increases implying
that the optimal level of consumption decreases; on the numerator level, as Ht
increases, labor productivity goes down, consumption decreases even more: the
OPT curve goes down in the C − H space.

Second, from the aggregate resource constraint and the production function, we
can close the model:
Zt F(K̄, Ht ) = Ct + Gt
This time, consumption and hours worked go in the same direction, implying that
this equation’s curve is going up in the C − H space. Since this equation uses
both constraints (budget as the aggregate and the production one), we call it the
additive (ADD) condition.

Together, ADD and OPT give us a nice graphical setting to analyze our simple RBC
model. Nevertheless, we need one last equation to finish the analysis: the Euler
equation. In fact, recall that the Euler equation gives a link between consumption
and savings, the exact link we used to express both labor and consumption in the
same space. This equation is therefore the closing point of the model. By varying
rt such that the consumption path is consistent without having to use bonds, the
Euler equation gives us the last element of our model.

Graphically, our model looks like this:

41
3.3 Dynamics and results

3.3.1 Shocks to G

Permanent positive shock to G

The effects of a positive permanent shock to G are all related to the downward
shift of the ADD curve:

42
This downward shift comes from the fact that setting higher taxes (increasing G)
force people to consume less for the same amount of hours worked. However,
since it is a lump-sum tax (no distortion), it does not affect the optimal share of
consumption and labor directly.

From the graph, it is straightforward to see that:

C ↓: The shift of the ADD curve implies a drop in consumption.

H ↑: Hours worked also increase due to the shift of ADD.

W ↓: Since hours worked increase and capital is fixed, labor productivity is

decreasing (FHH
00 < 0). In a fully competitive market, this means that wages

go down.

Y ↑: Since hours worked increase and capital is fixed, output is increasing (FH0 >
0).

X ↑: A higher level of labor increases the marginal productivity of capital, hence

increasing the return on the capital and dividends.

r ∼: A permanent change in G will not affect the relative path of consumption.

That means that the ratio Ct /Ct+1 will not change, hence no effect on interest
rates.

43
Temporary positive shock to G

The effect on impact of the shock will be the same as the permanent shock for
all variables except interest rates. As time goes by, all variables go back to their
previous steady-state growth.

The difference in the interest rates variation comes from the perturbation of the
consumption path. In fact, since consumption goes down on impact and comes
back to its steady-state, we know that until the new equilibrium, consumption
at t + 1 will be higher than at t. This fact implies that a PIH consumer will want
to smooth consumption by using future consumption today (i.e. borrowing).
However, there is no vehicle to do so in this economy, the interest rate has to
go up so that consumers do not want to borrow. We have that r ↑ following a
positive temporary shock to G.

3.3.2 Shocks to Z

Permanent shock to Z

A positive shock to Z will have two effects on the economy. First, it shifts ADD
Z F (K̄,H ) 1/σ
 0 
up as Ct = Zt F(K̄, Ht ) − Gt . Second, it shifts OPT up as Ct = Vt 0(HH̄−H )t . The
t
issue is that with two shifts in the same direction, we know that consumption
will go up, but the effect on hours worked is unclear. Before exploring how we
can make sense of this question, let’s look at the graph and at the effects that are
known:

44
C ↑: Both curves movements imply an increase of consumption for the same
leisure: hence, consumption goes up.

L?: The effects on the labor/leisure decision is indeterminate.

W ↑: Whatever the movement in hours worked, wages will increase following a

left shift of labor supply and a right shift of labor demand.

Y ↑: Even without information on the effect on hours, we can safely assume

that output will go up, since productivity icreases and the level of hours
cannot be overly negative nor positive.

X ↑: Again, the fact that the effect on H is not straightforward does not over-
power the increase in productivity, leaving X with a positive effect.

r ∼: A permanent change in Z will not affect the relative path of consumption.

That means that the ratio Ct /Ct+1 will not change, hence no effect on interest
rates.

Now, in order to try and guess what actually happens to hours worked, consider
writing both curves as functions of hours worked:

Ct + Gt Ctσ
ADD: = F(K̄, Ht ) OPT: = G(Ht )
Zt Zt

45
where F is the production function (an increasing function in H, and G is the ratio
of wage to disutility in labor, a decreasing function in hours worked. If you divide
both equations, you can study how the two effects work; for example, dividing
OPT by ADD, you get approximately (without the government expenditures):
G(Ht )
Ctσ−1 =
F(K̄, Ht )
If σ = 1, meaning in the case of log utility, Ctσ−1 = 1: both effects are equal,
hours worked do not vary. If σ < 1, then an increase in Ct will decrease the LHS,
implying a decrease in the RHS: hours worked have to increase. In the opposite
case, if σ > 1, then an increase in Ct will increase the LHS, implying an increase
in the RHS: hours worked have to decrease.

Temporary shock to Z

As studied in the government expenditures case, the effect on impact of a tem-

porary shock will be the same as the permanent shock for all variables, except
interest rates. However, here the consumption path will change in the opposite
direction: consumption goes up first, then goes down, meaning consumption
today will be higher than tomorrow until the new equilibrium. This means that
consumers will want to save (i.e. buy consumption bonds); but as we now know,
these bonds cannot be exchanged, therefore the interest rate has to decrease in
order to make them less attractive. We have r ↓.

3.4 Interpretation of the model

Summary of the results

The summary of the model and its effects is in the table below:

We get that in response to shocks on G, consumption and wages go down, while

output and hours worked react in the opposite direction. This means that they
are countercyclical following a G shock: this is not what we want to see from our
model in terms of replication of actual observations.

46
Following a Z shock however, all our variables seem to comove as we want them
to, except for hours worked, which can go both ways.

These results are not quite satisfying since they do not show the actual data. This
leads us to believe that the simple RBC model described here is (1) better for
interpretation of the technology shocks, (2) lacking some key ingredients to make
it more accurate. Moreover, one could argue that these effects are unconditional
responses (the effect is alone, not affected by anything else). Impulse responses
estimation would give a better result to show these conditional correlations. But
in order to estimate an IRF, you need to identify the shocks hitting the economy,
a hard task.

Comovement of C and L: the Barro-King problem

We have seen two types of real shocks under the simplest RBC model: G shocks
and Z shocks. The issue emphasized by Barro and King (1984) revolves around
the fact that in the data, H and C seem to comove always. As we’ve seen, it is not
the case following a shock on G and can be also missed by a shock on Z. What
can we do about it? We need to challenge the optimality conditions (OPT) such
that they do move together.

Technology shocks and labor supply

We saw that technology shocks are close to reproducing the effects we see in the
economy. But even if we consider only technology shocks, this simple RBC model
is not as satisfying as we would want it to be. Indeed, hours worked are subject
to both wealth and substitution effects, leaving it hardly variable following a Z
shock: the volatility of hours worked is not even close to output volatility. This
issue calls for an amplification of labor supply shocks, meaning a very procyclical
wage.

47
3.5 Appendix

3.5.1 Fixed labor and capital: partial equilibrium RBC

Suppose for the sake of it that labor is also supplied inelastically (as capital) like
we had in the growth models of the first semester. What would the previously
studied shocks deliver in terms of dynamics? Before trying to solve the model,
note that supplying labor inelastically is the same as having a fixed amount of
hours worked and of leisure, thereby fixing the marginal utility of leisure at V 0( L̄)
as well as the production function and its derivative (when you don’t account for
productivity).

Productivity shock

3.5.2 Capital shock

3.5.3 News shocks

A news shock is defined as a particular shock (G, Z or anything else) that is

expected to happen in the future. The dynamics following that type of shocks
are different since consumers have time to take action before the shock. Indeed,
since households are always looking to smooth consumption, any shock that has
a future effect on C will cause a current change on behavior. However, recall that
this model does not include capital stock, nor consumption bonds in equilibrium.
Hence, it is not possible for households to prepare for an event before it occurs.
In order to wrap up our minds around this idea, let’s use the example of a news
shock on technology Z: in one period, a permanent shock on Z will occur.

We already know what happens following a Z shock under certain parameters:

an increase in consumption and in output. In general, we can characterize this
shock as making households richer. A richer household will generally work less,
so labor supply shifts left immediately (even if the shock happens in one period).
In contrast, labor demand will not shift out right away since the productivity
gains have not yet happened. The combination of those effects in the labor market

48
will cause hours worked to decrease and wages to increase on impact. This causes
output to fall (hours worked fall while capital stock and productivity remain fixed).
Moreover, a household that knows it will be consuming more in the following
periods will want to smooth out this increase even today. But in this model,
households have no vehicles of savings to sell in order to finance consumption
now, they cannot increase consumption right away and the interest rate has to
rise to compensate.

49
Chapter 4

Advanced RBC models

As we advance into more complicated models, we need to discuss the full RBC
model. A full RBC model means that we will introduce variable capital stock
(investment) to our model, but we will also solve the model numerically and
simulate our way through shock analysis.

4.1 Model

4.1.1 Households

Assumptions

We use the typical model of household behavior, in which we allow for capital
(Kt ) accumulation via [Link] the only change to the problem will be
on the budget constraint.

We get our usual multi-period objective function:

"∞ #
Õ
max Et β u(Ct+s ) + V(H̄ − Ht+s )
s

Ct ,Ht ,Kt+1,Bt+1
s=0

50
in which we assume that u(Ct ) = ln(Ct ) for simplicity.

The budget constraint now allows for purchasing either consumption bonds
(zero in equilibrium) yielding a return of rt or capital stock with a return Rt that
depreciates at a rate δ. The household gets his income from previous periods
assets, labor income and profits, while he has to pay for his consumption and
taxes. Therefore the full budget constraint is written as:

Ct + Bt+1 + Kt+1 = rt Bt + [Rt + (1 − δ)]Kt + Wt Ht + Πt − Tt

This budget constraint can also be summarized as the national accounts identity:
Y = C + I + G where I represents investment in the capital stock.

Solution

The solution to this problem is fairly similar to what we have seen in the previous
chapter. Four FOCs are important:
1
= λt
Ct

V 0(H̄ − Ht ) = λt Wt
λt = Et [β(Rt+1 + (1 − δ))λt+1 ]
λt = Et [βrt+1 λt+1 ]
The last two equations are two sides on the same intuition: a Euler equation for
risk-free bonds and one for risky assets. We will use the risky asset one for the
actual Euler equation while transforming the risk-free condition into an arbitrage
condition:
λt+1 λt+1
   
Et β(Rt+1 + (1 − δ)) = Et βrt+1
λt λt
⇔ Rt+1 ≈ rt+1 + δ + Risk premium
And the Euler equation is given by:
 
Ct
1 = Et β(Rt+1 + (1 − δ))
Ct+1

51
4.1.2 Firms

Now we turn to the firm which is a very simple model in this RBC. As before,
the production function is a neoclassical production function in capital and labor,
augmented by technology Zt such that: Yt = F(Kt , Zt Ht ).

The firm’s problem is to choose its capital and labor inputs and its output as to
maximize profits in the following manner:

max Yt − Wt Ht − Rt Kt s.t. Yt = F(Kt , Zt Ht )

Yt ,Kt ,Ht

which can be solved by replacing the constraint in the objective function in order
to express it only as factor inputs. We get the following two FOCs:

FK (Kt , Zt Ht ) = Rt

Zt FL (Kt , Zt HT ) = Wt

4.1.3 Government shocks

As in the previous model, we assume government action is an exogenous process

Tt = Gt where Gt is the exogenous process subject to shocks, either permanent
or temporary. Note that government expenditures are not part of any utility or
profits, they are pure waste.

The shocks on G can be written in their dynamic form:

G̃t+1 = ρG · G̃t + εt+1

G
where Et εt+1 =0
 G 

4.1.4 Technology shocks

Technology is also assumed to be an exogenous process, following the process:

Z̃t+1 = ρ Z · Z̃t + εt+1

Z
where Et εt+1 =0
 Z 

52
4.1.5 Summary

Before going further into solving the model, this section provides a summary of
all equation and a small explanation on the interaction between them.

Household Optimization

1
= λt
Ct
V 0(H̄ − Ht ) = λt Wt
 
Ct
1 = Et β(Rt+1 + (1 − δ))
Ct+1

Firm Optimization

FK (Kt , Zt Ht ) = Rt
Zt FL (Kt , Zt HT ) = Wt

Constraints

Yt = Ct + It + Gt
Kt+1 = (1 − δ)Kt + It
Yt = F(Kt , Zt Ht )

Exogenous Shocks

G̃t+1 = ρG · G̃t + εt+1

G

Z̃t+1 = ρ Z · Z̃t + εt+1

Z

53
This sums up to a system of 10 non-linear equations that need to hold at the same
time. This model is not easy to solve, and with that level of complexity there
hardly is an analytical solution. Note that on the contrary to the previous model,
capital is not fixed so we cannot reduce the dimensionality of the problem: we
are now in a general equilibrium setting where the three markets (goods, assets,
labor) have to clear at the same time. Solving this model will make use of new
techniques developed in the following sections.

4.2 Solving the model

4.2.1 Log-linearization

Log-linearizing a system of equation allows to approximate a complex system

of non-linear equations as an easy system of linear equations defined in the
neighborhood of the steady state.

Taylor expansion method

Suppose Z = F(X) is the function we want to log-linearize around the steady-state

values Z ∗ = F(X ∗ ).

First, you take the logarithm on both sides:

ln(Z) = ln(F(X))

Then, you use the Taylor expansion of order one to the steady state value:

∂ ln(Z ∗ ) ∂ ln(F(X ∗ ))
ln(Z ∗ ) + (Z − Z ∗ ) ≈ ln(F(X ∗ )) + (X − X ∗ )
∂Z ∂X
1 1
∗
(Z − Z ∗ ) ≈ F 0(X ∗ )(X − X ∗ )
Z F(X ∗ )
Z̃ ≈ ε X X̃

where X̃ = ∆%X for any variable X and ε X is the elasticity of function F(·).

54
We clearly got a linear relation between X and Y , however, an unknown term
was added, namely the elasticity of function F(·). This new parameter will need
to be defined by the economist in a process called calibration.

Basu’s own method

This shortcut consists in taking logs and totally differentiating, starting from the
steady-state values. Again, let’s use the same function Z = F(X):

1. Start with the steady state values: Z ∗ = F(X ∗ )

2. Take the logarithm of the equation: ln(Z ∗ ) = ln(F(X ∗ ))
1 1
3. Totally differentiate: Z ∗ dZ
∗ ≈ 0 ∗
F(X ∗ ) F (X )dX
∗

x−x ∗
4. Using the fact that dx ≈ x − x ∗ and x̃ = x ∗ , find the log-linear formulation:
F 0(X ∗ ) X − X∗ ∗
Z̃t = · dX
F(X ∗ ) X − X∗
(X − X ∗ )
⇔ Z̃t = F 0(X ∗ ) · X̃t
F(X ∗ )
⇔ Z̃t = ε ∗X · X̃t

Case 1: the Euler equation

Taking logs is impossible in this context since the presence of an expectation

operator forbids to use any non-linear operator. We will have to start with totally
differentiating:
λt = Et [βλt+1 (Rt+1 + (1 − δ))]
dλt ≈ Et [β(R∗ + (1 − δ))dλt+1 + βλ∗ dRt+1 ]
λ∗ λ∗ R∗
 
dλt · ∗ ≈ Et β(R + (1 − δ))dλt+1 · ∗ + βλ dRt+1 · ∗
∗ ∗
λ λ R
λ̃t · λ ≈ Et β(R + (1 − δ))λ̃t+1 · λ + βλ R̃t+1 · R
∗ ∗ ∗ ∗ ∗
 

λ̃t ≈ Et β(R∗ + (1 − δ))λ̃t+1 + βR∗ · R̃t+1

 

55
We now have a linear equation in expectation. Using the fact that in the steady
state, R∗ + (1 − δ) = r ∗ = 1/β, we can simplify our last equation to get:

λ̃t ≈ λ̃t+1 + βR∗ · R̃t+1

Case 2: the Production function (CRS + PC)

We start with the assumption that the production function is of the form: Yt =
F(Kt , Zt Ht ). Additional assumptions include that the firm is under perfect com-
petition and that F(·) follows CRS.

This time, let’s start by taking logs and taking the total differential:

ln(Yt ) = ln(F(Kt , Zt Ht ))
F1 (K ∗, Z ∗ H ∗ ) F2 (K ∗, Z ∗ H ∗ )
dKt + dZt + dHt


⇔ Ỹt ≈
F(Kt , Zt Ht ) F(Kt , Zt Ht )
⇔ Ỹt ≈ εK K̃t + εW H ( Z̃t + H̃t )

and we can use the fact that CRS implies that the elasticity of input is equal to
the input share to write:
 ∗  ∗
WH WH
Ỹt ≈ 1 − K̃t + ( Z̃t + H̃t )
Y Y

where the parameter to calibrate will be the share of labor in the production.

56
Case 3: Labor demand

Now, let’s try with labor demand:

Zt · FL (Kt , Zt Ht ) = Wt
ln(Zt ) + ln(FL (Kt , Zt Ht )) = ln(Wt )
FLK (K , Z H )
∗ ∗ ∗ FLL (K ∗, Z ∗ H ∗ )
Z̃t + dKt + (dZt + dHt ) = W̃t
FL (K ∗, Z ∗ H ∗ ) FL (K ∗, Z ∗ H ∗ )
FLK (K ∗, Z ∗ H ∗ ) · K ∗ FLL (K ∗, Z ∗ H ∗ ) · (Z ∗ H ∗ ) Z̃t H̃t
Z̃t + K̃t + ( ∗ + ∗ ) = W̃t
FL (K , Z H )
∗ ∗ ∗ FL (K , Z H )
∗ ∗ ∗ H Z

4.2.2 Calibration

As we have seen, some variables are present in the log-linearized system without
having been defined by the the model’s condition. Hence, we must find values
for these parameters that imply a correct specification for the economy. There
are two solutions for this issue: calibration and estimation.

Estimation might be a more rigorous technique as it requires to use actual data

to find parameter values. However, following this path might lead to few irregu-
larities. First, estimation is a lot more difficult than calibration in the sense that
you have to find data, a good model, robust analysis, etc. Second, you might not
be able to estimate all parameters, depending on the data. In fact, the Minnesota
school started a tradition to use steady-state data (average shares, average interest
rates, etc.). The problem with this restriction is that it rules out some curvature
parameters that cannot be present in steady-state data. Trying to get around this
issue can also be problematic. For example, using micro-data may seem to be a
good solution but it is often the case that sets of data are too restricted to say
anything about aggregate behavior. Moreover, some might argue that there can
be great differences between microeconomic conclusions and their aggregation.

In the end, calibrating seems to be the easiest choice, even though it leads to
many strong debates in the profession. For now, we will keep it as is and talk
more in depth about it later.

57
4.2.3 System of log-linearized equations

In the end of that process, you should end up with a series of log-linearized
equations, with parameters to fit in. This system of equations will help you
solve dynamic stochastic general equilibrium models using Matlab (Dynare) or
anything else.

Let’s transform all our model equations into their respective log-linearized forms.

Optimal consumption path

1/Ct = λt ⇔ −C̃t = λ̃t

Labor supply

V 0(H̄ − Ht ) = λt Wt
V 00(H̄ − H ∗ )
⇔− dHt = λ̃t + W̃t
V 0(H̄ − H ∗ )
V 00(H̄ − H ∗ ) · (H̄ − H ∗ ) H ∗
⇔− 0 · dHt = λ̃t + W̃t
V (H̄ − H ∗ ) · (H̄ − H ∗ ) H ∗
V 00(H̄ − H ∗ ) · (H̄ − H ∗ ) H∗
⇔ − · · H̃t = λ̃t + W̃t
V 0(H̄ − H ∗ ) (H̄ − H ∗ )
| {z } | {z }
Elasticity of leisure marginal utility Ratio of hours to leisure

⇔ H̃t = εHW · (λ̃t + W̃t )

Euler Equation

(See 5.2.1, case 1):

λ̃t ≈ λ̃t+1 + βR∗ · Et R̃t+1

 

58
Capital demand

FK (Kt , Zt Ht ) = Rt
∗
F11 ∗
F12 · H ∗ ∗ · Z∗
F12
⇔ ∗ dKt + dZt + dHt = R̃t
F1 F1∗ F1∗
∗ · K∗
F11 ∗ · Z ∗H∗
F12 ∗ · Z ∗H∗
F12
⇔ K̃t + Z̃ t + H̃t = R̃t
F1∗ F1∗ F1∗
∗ · K∗
F11 ∗ · Z ∗H∗
F12
⇔ K̃t + · ( Z̃t + H̃t ) = R̃t
F1∗ F1∗

And since, from Euler’s theorem, F1 is of homogeneous degree 0, we can write:

∗
F11 · K ∗ + F12
∗
· Z ∗H∗ = 0
∗ ∗
⇔ −F11 K = F12
∗
· Z ∗H∗
which leads to the following simplification:
∗ · Z ∗H∗
F12
⇔ ( Z̃t + H̃t − K̃t ) = R̃t
F1∗
F∗ Z ∗H∗ · F∗ F∗
⇔ 2 ∗ ∗ 12∗ ( Z̃t + H̃t − K̃t ) = R̃t
F · F1 F2
1
⇔ sH · ( Z̃t + H̃t − K̃t ) = R̃t
εK H

Labor demand

(See 5.2.1, case 3):

sK
(K̃t − Z̃t − H̃t ) = W̃t
εK H

59
Aggregate resource constraint

Yt = Ct + It + Gt
⇔ Yt − Y ∗ = Ct − C ∗ + It − I ∗ + Gt − G∗
Yt − Y ∗ Ct − C ∗ + It − I ∗ + Gt − G∗
⇔ =
Y∗ Y∗
C ∗ I∗ G∗
⇔ Ỹt = C̃t · ∗ + I˜t · ∗ + G̃t · ∗
Y Y Y
˜
⇔ Ỹt = C̃t · sC + It · (1 − sC − sG ) + G̃t · sG

Law of motion of capital

The law of motion of capital states that:

Kt+1 = (1 − δ)Kt + It

and in the steady state, K ∗ = (1 − δ)K ∗ + I ∗ ⇔ I ∗ = δK ∗ . Hence,

Kt+1 = (1 − δ)Kt + It
⇔ Kt+1 − K ∗ = (1 − δ)Kt − K ∗ + δK ∗ + It − I ∗
(Kt − K ∗ ) I ∗ ˜
⇔ K̃t+1 = (1 − δ) + ∗ It
K∗ K
˜
⇔ K̃t+1 = (1 − δ)K̃t + δ It

Production function

(See 5.2.1, case 2):

 ∗  ∗
WH WH
Ỹt ≈ 1 − K̃t + ( Z̃t + H̃t )
Y Y

60
Shocks

G̃t+1 = ρG · G̃t + εt+1

G

Z̃t+1 = ρ Z · Z̃t + εt+1

Z

4.3 Dynamics and results

4.3.1 Shocks to Z

Assume a shock to technology (productivity) Zt . At first, we are not going to say

anything about the persistency of the shock but we will see that at some point
we will have to.

From our system of log-linearized equations, a shock to technology has a direct

effect on:

• Capital demand: a higher productivity means that capital stock will yield
a greater marginal return, hence an increase in its price (perfect competition
assumption).
1
sH · ( Z̃t ↑ +H̃t − K̃t ) = R̃t ↑
εK H
• Labor demand: in the same way, a higher productivity means that labor
will yield a greater marginal return, hence an increase in its price (perfect
competition assumption).
sK
Z̃t + (K̃t − Z̃t − H̃t ) = W̃t
εK H
sK sK
(1 − ) Z̃t ↑ + (K̃t − H̃t ) = W̃t ↑
εK H εK H
• Production: however, since the production function is not an optimal
equation but rather a constraint, let’s leave it for later.

These effects increase current and future interest rates and wages which will also
affect households in their wealth. While it could any way, our parametrization

61
assumes that consumption will go up on impact. This effect is the main driver for
determining the rest of the variables later.C ↑

Equilibrium in the labor market

Now we turn to the labor market. Since we know that following a productivity
shock we have a positive wealth effect, we can assume that labor supply will
shift to the left. We already know that labor demand will go to the right. These
two effects together will have the combined effect of rising wages (certainly) but
the effect on hours worked in uncertain. Again, we use our parametrization to
determine the direction of the effect: hours worked increase. W ↑ H ↑

Equilibrium in the capital market

The capital market is fairly easy to analyse on impact. We know that capital
demand increased and since capital supply is fixed on impact (it takes one period
to create capital), the interest rate will have to increase. R ↑

Investment and Output

Finally, we can determine what happens to output and investment. Regarding

investment, we know that households see an increase of consumption right on
impact, while if the shock is transitory, they will want to smooth out this increase
in wealth over periods, by investing (saving). This means that investment will go
up on impact. From the resource constraint, we should observe an increase in
output. Moreover, hours worked increased as well as productivity and capital will
increase progressively. It is therefore now clear what happens to output: output
should go up! I ↑ Y ↑

Impulse response functions

Now that we know what happens on impact, we can try and evaluate what are
the potential dynamic effects in order to draw impulse response functions.

62
Since investment is up, the economy will experience capital accumulation for
some periods (the capital stock will increase). This dynamic is not going to
last as with time, less and less smoothing is required and both investment and
consumption will start declining. An increasing capital stock will cause labor
productivity to continue to rise and hence will shift labor demand to the right,
pushing wages up. This effect will be higher as consumption continues to grow
in the beginning but will be slowly decreasing once consumption starts going
down. The effect on hours is still ambiguous but our estimated IRFs show that
hours worked will go down. From the capital market, since capital supply will
start to catch up (to the right), interest rates will slowly go down.

63
4.3.2 Shocks to G

Since government expenditures do not enter in neither household utility nor firm
profits, a positive shock to G is not going to produce great shocks to the optimal
allocations between variables. Nevertheless, G has a direct negative effect on the
budget constraint, through taxes. This causes a negative wealth effect, forcing
consumption to decrease on impact. Investment will decrease by even more to
counteract the decrease of consumption. C ↓ I ↓

Equilibrium in the labor market

Consumers experience a loss of wealth, pushing them to work more for the same
wages (to catch up for the loss), hence labor supply shifts to the right: hours
worked increase and wages decrease. Labor demand is not affected on impact
since G has no effect on the firm and even though investment has gone down, it
takes one period for capital to adjust and decrease. H ↑ W ↓

Equilibrium in the capital market

Nothing has changed in this market since neither capital supply (fixed on impact)
nor capital demand (firm optimization) have changed on impact. R ∼ K ∼

Output

We have seen that both consumption and output will fall following a positive fiscal
shock. This means that the effect on output we can observe from the resource
constraint is unclear (C,Y go down, G goes up). However, we can see from
employment that as hours worked go up, it should be the case that output goes
up. Y ↑

64
Impulse Response Functions

Following the decrease of investment and consumption, and as the budget con-
straint progressively becomes less tight, we will see upward movements of both
variables back to the steady state. Meanwhile, the fall in investment consumes
some of the capital stock accumulated leaving K to fall as well. A decrease in
capital stock has a negative effect on labor productivity and a positive effect on
capital productivity. Hence, labor demand will shift out, and with labor supply
shifting back, we will see wages going back up and hours going down. Capital
productivity going up means a shift of capital demand to the right while capital
supply goes to the left: we can expect an increase in the interest rate.

65
4.4 Interpreting the results

In order to assess the efficiency of the RBC model, one must define criterias against
which it is possible judge the output of the model. Cooley and Prescott (1995) are
among the earliest to do this and they choose to match relative volatilities of key
moments.

For instance, take a one-time unique technology shock. The benchmark RBC
model will predict:

• σI > σY > σC : that is, investment is the most volatile component while
consumption is relatively smooth in comparison to output.

• Variables Y, I, C and H will commove following a shock.

However, the model has some inconsistencies with the data. In particular, the
model does not generate enough volatility of interest rates. Further, it generates
wages and real interest rates that are far too procyclical relative to the data. In the
data, wages are very modestly procyclical and real interest rates are acyclical or
countercyclical, depending on how you measure them. Moreover, the variations
of C and H are too smooth compared to actual observations. As a result, labor
productivity will be too much correlated with output (since H doesn’t change),
also something that is not observed. It also seems that r, the interest rate is
too strongly procyclical. Finally, at a deeper level, people have criticized RBC
models because they don’t seem particularly realistic. To generate fluctuations
that resemble those in the US, one needs large, high frequency variation in Zt .
The issue here is that a technology shock must be present to move the model,
but it seems highly unprobable to have a negative shock to technology to model
recessions.

After thorough analysis, it seems that the RBC model makes a lot of mistakes
when compared to actual observations. The first counterargument to this apparent
issue is to ask if we can really compare the shocks fed into the RBC model to
shocks happening in real life. In fact, RBC shocks are unique and unconditional,
meaning nothing else happens at the same time. We have studied that some
shocks (in particular shocks on G) can affect consumption the other way: could it
be that real life shocks are affected by different contradictory shocks at the same
time? Probably. Hence, we should theoretically feed the model with the same mix

66
of shocks that happen in the data: this is very difficult to do.

Critics of the real business cycle model are uncomfortable with the facts that
it is (a) driven by technology “shocks” and that (b) these shocks must be large
and sometimes negative. Hence, much of business cycle research since the 1980s
has been involved in modifying the basic model to (a) allow other shocks to
“matter” in a way that they can’t in the basic model (e.g. monetary policy) and (b)
generating better and more realistic mechanisms for the model to take “small”
shocks (as opposed to large) and produce relatively large business cycles.

4.5 Appendix

4.5.1 Capital Shocks

4.5.2 Patience shocks

In the models we described since the beginning, we have used the parameter
1
β ≡ 1+ρ to represent the “patience” of households. In particular, a high β means
that households value future utility very much (i.e. they are patient) while a low
β means the exact opposite. We will see later that patience can be very important
to model as it is a defining characteristic of the quantity of goods a households
wants now instead of later. That’s why we look at this type of shocks in the case
of our second RBC model (with capital).

Let’s assume a negative shock to β, meaning households would rather consume

more now than they used to. From the Euler equation, we can see that this has a
clear negative effect on the current Lagrangian multiplier λt meaning that this is
equivalent to a positive wealth effect: people feel richer. This means that:

• Current consumption goes up.

• Labor supply shifts left: hours worked decrease and wages increase.

• Output decreases as hours worked decreases.

• Consumption goes up while output goes down: investment must go down.

67
4.5.3 Laziness shocks

In this particular case, suppose that marginal utility derived from leisure increases.
In order to see that in an easier manner than with the general V(H̄ − Ht ), let’s
assume that utility from leisure takes the following form:

V(H̄ − Ht ) ≡ θ ln(H̄ − Ht )

Then a shock to laziness could be modeled as a shock to the parameter θ.

When this happens, you could compare it to increased laziness of households

and hence it will cause a shift to the left to the labor supply, without affecting
labor demand: wages go up and hours worked go down. We can also argue from
the combination of all FOCs that consumption will decrease as well. Moreover,
if hours worked decrease, then output decreases as well since productivity and
capital stock didn’t change. In general, this shock functions as a negative wealth
effect, which will force households to decrease investment in order to smooth
consumption. Finally, since the capital stock slowly decreases after impact, the
interest rate will increase.

4.5.4 News shocks

As we have seen in the previous chapter, news shocks work like wealth shocks in
the sense that households expect what is going to happen in the future. Let’s use
the example of a positive news shock.

Households feel richer and hence will:

• Increase current consumption (because of smoothing)

• Reduce their labor supply (they don’t need to work so much anymore)

• Reduce investment (again because of smoothing)

This will additionally cause output to go down, wages to go up, interest rates to
go up progressively.

68
4.5.5 Uncertainty shocks

69
Chapter 5

Non-Walrasian RBC models

5.1 Externalities in production

This chapter will uncover some of the effects implied by extreme marginal products
in production. The rationale behind this theory is to try and get out of the result
that only a shock in Z will make Y, C, I and H comove. We could also use the
rationale that facts are not matched yet and we would want to find other theories
that could potentially lead to better matching models. In order to do this, we’ll
introduce increasing marginal products via Marshallian externalities: the classical
assumption that when firms produce more, other firms get more productive.

5.1.1 Model

Firms

As we will see, the main difference between this model and the previous one
is how firms are modeled. In particular, instead of having one representative
firm, we will have a continuum of firms, indexed by i where i ∈ [0, 1]. Each firm
will have the same production technology (function) but will produce a different
output, denoted Yit (output produced by firm i at time t). The production function

70
is given by:
Yit = At · Kitα (Zt Hit )1−α
The main difference between this production function and the one in previous
models is that At is a new variable, called the externality. As you can see, At is not
so different than Zt in the firm’s view as it is an exogenous, time-varying input.

Nevertheless, At will play a different role in the whole economy as it is in fact

an endogenous variable for the economy as whole. In fact, we define At as an
increasing function of aggregate output Yt : by producing more firms get more
productive, but they do not internalize this process. Formally,

1  1− γ1
1− 1
∫
At = Yt γ = Yit di
0

The aggregate stock of capital Kt and labor Ht are defined in the same way
as aggregate output. From that aggregation, the whole economy’s production
function will be slightly modified:
∫ 1 ∫ 1 ∫ 1
α 1−α 1− 1
Yt = Yit di ⇔ Yt = At Kit (Zt Hit ) di ⇔Yt = Yt γ Kitα (Zt Hit )1−α di
0 0 0
∫ 1
1− γ1
⇔Yt = Yt Kitα (Zt Hit )1−α di
0
1
∫ 1
⇔Yt γ = Kitα (Zt Hit )1−α di
0
∫ 1 γ
α 1−α
⇔Yt = Kit (Zt Hit ) di
0
γ
⇔Yt = Ktα (Zt Ht )1−α


The individual firm’s problem is now:

max At Kitα (Zt Hit )1−α − Wt Hit − Rt Kit

Kit ,Hit

yielding the following two FOCs (or input factor demands):

αAt Kitα−1 (Zt Hit )1−α = Rt

and (1 − α)At Kitα Zt1−α Hit−α = Wt

71
   γ  1− γ1 α(γ−1)
We know that At = Ktα (Zt Ht )1−α = Kt (Zt Ht )(1−α)(γ−1) . Moreover,
in equilibrium, it must be that all firms have the same level of inputs (since the
prices are the same for all firms and they are price-takers in input markets). This
means that Kit = K jt for any i, j. Therefore,
∫ 1 ∫ 1
Kt = Kit di = Kit 1di = Kit
0 0

and the same applies for labor, Ht = Hit .

Consequently, we have a new formula for labor demand:

α(γ−1)
Wt = (1 − α)Kt (Zt Ht )(1−α)(γ−1) Ktα Zt1−α Ht−α
αγ (1−α)γ (1−α)γ−1
= (1 − α)Kt Zt Ht

which in turn gives a new log-linearized equation:

W̃t = αγ K̃t + γ(1 − α) Z̃t + [(1 − α)γ − 1]H̃t

And a new formula for capital demand:

α(γ−1)
Rt = αKt (Zt Ht )(1−α)(γ−1) Ktα−1 (Zt Ht )1−α
αγ−1
= αKt (Zt Ht )(1−α)γ

which in turn gives a new log-linearized equation:

R̃t = (αγ − 1)K̃t + (1 − α)γ[ Z̃t + H̃t ]

These two input factor demands may not seem different than the usual ones,
however the externality parameter γ holds an important change: their slopes
are now linked to γ. For example, when γ increases (externalities increase) the
labor demand curve gets flatter and flatter, until it reaches positive territory:
labor demand is sloping up! A positive slope of labor demand would give new
interpretations for shocks, but it could also lead to non-convergent steady-states,
ultimately ruining our model. We must first make sure that there is in fact a
steady-state around which we log-linearize. The requirement for an existing
steady-state is that γα < 1 ⇔ γ < 1/α ⇔ γ < 3. For labor demand to slope up

72
 1
we need (1 − α)γ − 1 > 0 ⇔ 1−α < γ ⇔ γ > 3/2. We’ll therefore look at effects
under values of γ lying between [1.5, 3).

Finally, let’s log-linearize the aggregate production function:

γ
Yt = Ktα (Zt Ht )1−α


⇔ Ỹt = γα K̃t + γ(1 − α)( Z̃t + H̃t )

Rest of the model

In essence, the rest of the model is our typical RBC model so we are not going to
detail the computations here. However, a summary of the system of log-linearized
equations is provided below:

Optimal consumption path:

−C̃t = λ̃t

Labor supply:
H̃t = εHW · (λ̃t + W̃t )

Euler Equation:
λ̃t ≈ λ̃t+1 + βR∗ · Et R̃t+1
 

Aggregate resource constraint:

Ỹt = C̃t · sC + I˜t · (1 − sC − sG ) + G̃t · sG

Law of motion of capital:

K̃t+1 = (1 − δ)K̃t + δ I˜t

Shocks:

G̃t+1 = ρG · G̃t + εt+1

G

Z̃t+1 = ρ Z · Z̃t + εt+1

Z

73
5.1.2 Results

For the results of this model, keep in mind that the reasoning is approximately
the same as in classical RBC models, only with an upward-sloping demand for
labor. This will help us to identify easily the quantitative differences that are
caused by externalities. To do that, we consider three models: the classical RBC
(no externalities), a moderate level of externalities and high externalities.

Shocks to G

A shock to G is essentially a negative wealth effect, pushing labor supply down. In

the classical RBC model, this caused an increase in hours worked and a decrease in
wages. With externalities however, labor demand is sloping up, which translates
into higher wages in addition to higher hours worked. This could even mean
that consumption could rise. In fact, in the RBC model, income does not change
that much since wages and hours move in different direction and overall taxes
increase. In the models with externalities, the common increase in both hours
and wages could potentially offset taxes increase. As we can see from the graphs
below, a moderate level of externalities might not be enough, while a high level
would do the work.

We also know that higher number of hours will mean higher output, an effect that
will be emphasized as externalities are more and more important, and consumption
suffers less and less from taxes. The direction of consumption will also change
what happens with investment. It will be going down when consumption goes

74
down but up in the cases with externalities. The direction of consumption is the
same in all cases and therefore interest rate will go up on impact and go down.

C ↑; H ↑; W ↑; Y ↑; I ↑

You can see the IRFs of the aforementioned shock below. Note that the blue line
shows the RBC case, while green and red lines show the effects under externalities.

75
Shocks to Z

A shock to Z will have: (1) a positive wealth effect for households and (2) a
positive effect on input marginal returns for firms.

Therefore, in the labor market, labor supply shifts to the left while labor demand
shifts to the right. Since labor demand is now sloping up, this creates increasing
wages and hours worked (less ambiguous than in classical RBC models). The
increase of hours worked also means an increase in output.

In the capital market, the increase in productivity implies higher capital demand,
thus a higher interest rate. Investment goes up as households smooth consump-
tion, meaning that capital stock increases progressively, bringing back the interest
rate to its steady-state.
C ↑; H ↑; W ↑; Y ↑; I ↑
You can see the IRFs of the aforementioned shock below. Note that the blue line
shows the RBC case, while green and red lines show the effects under externalities.

76
5.1.3 Interpretation

The main results to draw from this new model is how externalities allow for
commovement oh C and H, even when the shocks are not technologically driven!
The key to that commovement relies on endogenous shifts in labor demand at the

77
firm level. This is very interesting to both reasons we delved into the model in the
first place. However, there seems to be very little evidence that in fact those kind
of externalities happen in the data. Therefore we might look for other ways to
implement the same type of endogeneity of firm labor demand in a new manner.

5.2 Imperfect Competition: the Rotemberg-Woodford

model

This section’s topic is imperfect competition and its implications for the typical
DSGE models. Imperfect competition is important since:

• It allows for more realism (describes actual situations where firms indeed
hold some market power).

• It allows to study situations in which at the firm level there are increasing
returns (see previous chapter).

• It allows non-technological shocks to have desired effects on the model

(think demand shocks or even sunspot shocks).

• It has different predictions as to the effects of technological shocks.

5.2.1 Model

Demand for differentiated goods

This model assumes a different kind of aggregation for products. We’ve seen in
the previous that although firms produce different products, they all sold at the
same price. This time we relax this assumption.

Let there be a continuum of differentiated commodities indexed by i ∈ [0, 1] such

that a unique firm produces a unique good that other firms cannot match. Those
goods are somewhat substitutable. The aggregate demand in the economy (C, I

78
and G) is for the composite good Y where
∫ 1  σ/(σ−1)
Yt = yit(σ−1)/σ di
0

In this equation, the new parameter σ is the price-elasticity of demand. It is

required to be strictly greater than 1.

All users of output solve:

∫ 1 ∫ 1  σ/(σ−1)
min pit yit di s.t. Yt = yit(σ−1)/σ di
yit 0 0

This problem has the following FOC:

∫ 1  1/(σ−1)
pit − ψt yit−1/σ yit(σ−1)/σ di =0
0
  −σ ∫ 1  σ/(σ−1)
pit
⇔ yit = yit(σ−1)/σ di
ψt 0
  −σ
pit
⇔ yit = Yt
ψt
And we can solve for Yt in:
# σ/(σ−1) # σ/(σ−1)
1 1−σ 1
"∫   " (σ−1)/σ ! ∫
pit Yt
Yt = Yt(σ−1)/σ
di ⇔ Yt = 1−σ
pit1−σ di
0 ψ t ψt 0
 ∫ 1  σ/(σ−1)
Yt 1−σ
⇔ Yt = p di
ψt1−σ 0 it
(σ−1)σ
∫ 1  σ/(σ−1)
1−σ
⇔ ψt (1−σ)
= pit di
0
∫ 1  1/(1−σ)
⇔ ψt = pit1−σ di = Pt
0

Which gives our final demand that the producer faces:

  −σ
pit
yit = Yt
Pt

79
Firms

Let each firm be a monopoly for the good market that it produces. Then, the
optimization problem of the firm is:

max pit yit − Wt hit − Rt kit s.t. yit = F(kit , Zt hit ) − Φ

pit ,hit ,k it

where h, k are firm-level inputs and Φ is a fixed cost. We can replace yit by the
demand function derived earlier. This yields the following Lagrangian:
  −σ    −σ 
pit pit
L ≡ pit Yt − Wt hit − Rt kit + φit F(kit , Zt hit ) − Φ − Yt
Pt Pt
the first-order conditions are:

• Price:  −σ !
pit−σ−1

pit
(1 − σ) Yt − φit (−σ) Yt = 0
Pt Pt−σ

⇔ (1 − σ) = φit (−σ)pit−1
σ
pit = φit ·
σ−1
• Labor:
Wt = φit Zt F2 (kit , Zt hit )

• Capital:
Rt = φit F1 (kit , Zt hit )

You can observe that φit , the Lagrangian multiplier of the production constraint,
can be interpreted as the cost of producing one more unit of output at the equi-
librium. In other words, it is the marginal cost of the firm at the optimal level.
σ
Now, because the price is a function of the marginal cost, we will define σ−1 as
the markup, denoted µ. We can hence simplify our system of FOCs into:
1
φit = pit
µ
µWt = pit Zt F2 (kit , Zt hit )
µRt = pit F1 (kit , Zt hit )

80
This formulation makes it clear that, since all firms are price-takers on the input
markets, they must have the same input levels as well as the same price: this is a
symmetric equilibrium. Therefore, as in the previous model:

pit = Pt ; yit = Yt ; kit = Kt ; hit = Ht

We can therefore normalize the price level to 1 to get:

1
φt =
µ
1
Wt = Zt F2 (kit , Zt hit )
µ
1
Rt = F1 (kit , Zt hit )
µ

Log-linearizing the new equations

In order to solve this new model, we need to log-linearize the four new equations
we got.

This is a fairly easy task for the marginal cost and input factor demands as not
much has changed. In fact, the marginal cost does not change at all, and since µ
is fixed, the two input demands have not changed either.

W̃t = α[K̃t − H̃t ] + (1 − α) Z̃t

R̃t = (1 − α)[H̃t + Z̃t − K̃t ]

For the production function, we have that Yt + Φ = F(Kt , Zt Ht ), by taking logs

and totally differentiating you get:

Yt + Φ = F(Kt , Zt Ht ) ⇔ ln(Yt + Φ) = ln(F(Kt , Zt Ht ))

1
⇔ ∗ dYt = sK K̃t + (1 − sK )( Z̃t + H̃t )
Y +Φ
Y∗
⇔ ∗ Ỹt = sK K̃t + (1 − sK )( Z̃t + H̃t )
Y +Φ

81
This form helps us see an interesting effect of fixed cost on production. Assume
∗
output grows by one dollar (Ỹt ↑), growth of inputs has increased by Y ∗Y+Φ dollars,
which is less than one. We get increasing returns to scale. We define γ ≡ Y Y+Φ
∗
∗ in
order to emphasize on the similarity with the previous model.

Because γ represents returns to scale, we can also write it as:

Average Cost C(·)/Y P · C(·)
γ= = 0 = 0 = µ(1 − sπ )
Marginal Cost C (·) C (·) · PY
where µ is the markup (as defined above) and sπ represents the profit share in
total revenues. In the steady-state, profits should be equal to 0, leaving us with
the fact that γ = µ in the steady-state. We can therefore write:
Ỹt = µα K̃t + µ(1 − α)[ Z̃t + H̃t ]

Rest of the model

In essence, the rest of the model is our typical RBC model, but there is a single
main difference: potential profits. Indeed, remember that the budget constraint of
the representative household includes a profit term Xt or Πt . Usually, this term is
zero always since firms are perfectly competitive but in this model, firms might
make some profits out-of-equilibrium. Indeed, recall that increasing inputs will
create more output than a one-to-one until we come back to the steady-state. This
fact is important to keep in mind as it can introduce shocks to markups.

As before, we are not going to detail all the computations of the rest of log-
linearizations here. However, a summary of the system of log-linearized equations
is provided below:

Optimal consumption path:

−C̃t = λ̃t

Labor supply:
H̃t = εHW · (λ̃t + W̃t )

Euler Equation:
λ̃t ≈ λ̃t+1 + βR∗ · Et R̃t+1
 

82
Aggregate resource constraint:
Ỹt = C̃t · sC + I˜t · (1 − sC − sG ) + G̃t · sG

Law of motion of capital:

K̃t+1 = (1 − δ)K̃t + δ I˜t

Shocks:

G̃t+1 = ρG · G̃t + εt+1

G

Z̃t+1 = ρ Z · Z̃t + εt+1

Z

5.2.2 Results

We have seen that the Rotemberg-Woodford model developed above is not so

different from the classical RBC model but in two elements: output is increasingly
affected by changes in inputs, wealth effects following positive shocks are higher
for households. Keeping in mind those two main differences, it should be fairly
easy to come up with the new conclusions of this model.

Shocks to G

We know that a shock to G is a negative wealth effect for households. However,

since wealth effect are now always higher, it will reduce consumption by less
if the overall effect of G is positive in inputs. And in fact, recall from the labor
market dynamics of the RBC model that, as labor supply shift right, hours worked
increase. Hence, we get
C ↓; H ↑; W ↓; Y ↑
Since consumption has fallen and will recover, investment will be going down as
well (smoothing).

In the end, the difference with the RBC model is not striking with regards to a G
shock: we just have more positive effects and less negative ones.

83
Shocks to Z

A shock to Z has (as before) two main effects: (1) a positive wealth effect (enhanced
by profits) and (2) a positive effect of marginal returns from inputs. This leads
to exactly the same effect as the RBC in all cases except for hours worked. In
fact, recall that in the RBC model, the effect on hours is ambiguous but is slightly
positive from our parametrization; in this model, the wealth effect pushes labor
supply more than in the RBC and can reduce hours worked for some values.

5.2.3 Time-varying markups

Exogenous markup variations

σ
Let us model stationary shocks to the price-elasticity of demand σ. Since µ = σ−1 ,
we will see changes in our markup as well. This shock will have effects on the
labor demand:
sK
W̃t = Z̃t + [K̃t − Z̃t − H̃t ] − µ̃t
εK H
but not on the aggregate output since

Ỹt = µ(1 − sH )K̃t + µsH ( Z̃t + H̃t )

depends only on the steady state value of µ.

We get the following IRF from the model:

We see that this kind of shocks create commovements of Y, C and H as well as

variances that are approximately what we would see in the data. Moreover, we
need only a value of γ = 1.2 in order to see those effects: way closer to reality
than γ = 2 from the externalities model.

One of the main issues though, would be the difference in labor productivity
which is quite acyclical while wages are strongly procyclical. Also, persistent
shocks in markups are the only type that make a lasting effect on the economy,
but is this kind of shock the one we observe in real life?

84
Endogenous markup variations

Now that we have seen that markups can have interesting effects, we want to dig
deeper. Of course, exogenous markup shocks are observed (anti-trust regulations,
etc.) but endogenous variations might be more frequent and potentially have
more effects on the economy. Therefore, we’ll specify a model for those markups
to change.

Let the elasticity of demand σ be a function of output such that:

σ(Yt )
µ(Yt ) = ⇔ µ̃t = ε µỸt
σ(Yt ) − 1
We’ll assume that price-elasticity rises with output, implying that a higher output
leads to more competition and lower markups. This gives a ε µ that is negative
(i.e. countercyclical markups). Parameterizing our model and analyzing the IRF
will help us determine if that is indeed the case.

Implications of markup variations

We have seen that our business cycle models allow for variation in efficient level
of output (per person) over time. However, we might ask ourselves if those
deviations are inefficient around the efficient level. Another way of stating the
questions would be: are there output gaps?

Markup variations can tell us about this phenomenon. For example, if µ is

procyclical, then fluctuations would be hindered by rent seeking from firm. Hence
fluctuations would be inefficiently small. However, if µ is countercyclical, then
the opposite happens and fluctuations are inefficiently large. This issue is why
we need to focus on measuring markup variations in neo-keynesian models.

Markup cyclicality

We have seen that under imperfect competition, the role of markups is very
important in the implications of the model. In particular, a main determinant of
these implications is the cyclicality of the markups (i.e. are markups pro-cyclical

85
or countercyclical?). How can we determine the answer to this question, and
what does it change?

Recall that markup is by definition the ratio of prices to marginal costs. Prices
are easily observed or estimated (price indices, etc.) and at the aggregate level
they can be normalized to 1. Hence, thinking about markup cyclicality is really
about cyclicality of marginal costs. Marginal cost can be decomposed as:
Pj
MC =
Fj

where P j is the price of input K, L or M, and Fj is the marginal product.

86
Chapter 6

New Keynesian models

New-Keynesian models are extensions of RBC models with nominal rigidities.

These nominal rigidities allow for a wide range of monetary phenomenons, ex-
ogenous or not, to have a real effect on the economy. A NK model usually has the
same backbone as RBC models with imperfect competition, without capital (first
two models) then with capital and with money and prices playing an important
role. Money is entering in a ”market” (exogenous money) or via a monetary policy
rule (Taylor rule).

6.1 NK model without K, flexible prices

6.1.1 Model

Households

In this model households solve two problems: one aggregate problem which
allocates consumption, labor and bonds (the one we know) and a consumption
problem in which they choose what type of good they purchase.

The first problem is almost identical to the RBC-type consumption model, with

87
the addition of prices:
1−σ 1+η
" ∞
#
Õ Ct+s H
max Et β s
− χ t+s
Ct ,Ht ,Bt
s=0
1−σ 1+η
" ∞  #
Õ Wt+s Bt+s−1 Bt+s
s.t. Et β s λt+s Ht+s + (1 + it+s−1 ) + Πt+s − Ct+s −
s=0
Pt+s Pt+s Pt+s
yielding the following FOCs:

• Consumption:
Ct−σ = λt

• Hours worked:
η Wt
χHt = λt
Pt
• Consumption bonds:
1 1
 
λt = Et λt+1 · β(1 + it ) ·
Pt Pt+1
 
Pt
⇔ Ct = β Et Ct+1 · (1 + it ) ·
−σ −σ
Pt+1
You can see that these equations are close to identical to their equivalent in the
RBC model.

The second problem that households are facing is to decide what allocation
differentiated products cit they should choose, given the optimal aggregate level
of consumption Ct (note that we use small caps to denote variables that differ
with respect to i: it is somewhat clearer that using only the index it ). We assume
aggregate consumption is given by:
θ
∫ 1 θ−1
 θ−1
Ct = cit di
θ

so that households solve the following problem:

θ
∫ 1 ∫ 1 θ−1
 θ−1
min pit · cit di s.t. Ct = cit di
θ
cit 0 0

88
This yields the following demand function:
  −θ
pit
cit = Ct
Pt

This demand function specifies the level of consumption of a particular good

as a function of its relative price and of total consumption. As in the previous
models of imperfect competition, each good has a single producer (monopolistic
competition), meaning that the level of prices a producer chooses will determine
his demand. The next section further explains the firm’s problem.

Firms

In a goods market equilibrium, we have that yit = cit : production from the firm is
equal to the amount asked by consumers. We can therefore rewrite the previous
demand function as: yit = (pit /Pt )−θ Yt . Moreover, assuming the production
function of an individual firm (with fixed capital stock) is given by: yit = Zt hit ,
the problem of the firm can be written as:

max L ≡ pit yit − Wt hit + φit [Zt hit − yit ]

pit ,hit
  −θ "   −θ #
pit pit
= pit Yt − Wt hit + φit Zt hit − Yt
Pt Pt

where φit is the Lagrangian multiplier of the production constraint: the shadow
price of producing one more unit, or equivalently, the nominal marginal cost
at the optimal level of production. Note that everything in this problem is in
nominal terms. This is a particularity of all New Keynesian models and this will
help us understand new dynamics once we introduce sticky prices.

The problem has two FOCs. In the pricing first-order condition, we get:
 −θ
pit−1−θ

pit
(1 − θ) Yt + φit · θ Yt = 0
Pt Pt−θ
1
⇔ φit · θ · = θ−1
pit

89
 θ
⇔ φit · = pit
θ−1
This can be interpreted in many ways. The previous form shows that price pit
is equal to nominal marginal cost times a (higher than one) constant called the
markup. We denote it µ ≡ θ/(θ − 1). If you twist the equation slightly, you get
that the real marginal cost is equal to the inverse of the markup:
φit
= 1/µ
pit
Since the markup is constant across firms, we know that all firms have the same
real marginal cost for a given markup.

For the demand in hours, the FOC is given by:

Wt
−Wt + φit Zt = 0 ⇔ = φit
Zt
This equation does not tell us much as it is right now, but we can set it in real
terms by dividing by pit to get:
Wt /pit φit
= = 1/µ
Zt pit
This tells us that the ratio of real wages to marginal productivity of labor is also
constant in this model. Finally, multiplying by Zt on both sides gives the labor
demand equation:
Wt Zt
=
pit µ

From these two conditions, we can see that φit and pit are pinned down by factor
prices and marginal productivity constant to all firms. This means that all firms
have the same nominal marginal cost and hence will choose the same price pit .
We have φit = φt and pit = Pt , so that we can rewrite our two FOCs as:
φt
= 1/µ and Wt /Pt = Zt /µ
Pt
This also has consequences on the choices of the firms, all firms choose the same
number of hours hit , thus producing the same output, the aggregate production
function can be written as:
Yt = Zt Ht

90
Goods market equilibrium

The goods market equilibrium condition is given the optimal consumption path
(the Euler equation) and the aggregate resource constraint:
 
Pt
Ct = β Et Ct+1 · (1 + it ) ·
−σ −σ
and Yt = Ct
Pt+1
This gives the IS curve of the New-Keynesian model (NKIS):
1 + it
   
Pt
Yt = β Et Yt+1 · (1 + it ) ·
−σ −σ
= β Et Yt+1 ·
−σ
Pt+1 1 + πt+1

By log-linearizing around a zero-inflation steady-state you get:

 1 
i˜t − Et [πt+1 ]

Ỹt = Et Ỹt+1 −

σ
This log-linearized version of the NKIS shows the demand side of the economy (it
can be viewed as the aggregate demand curve of intermediate macroeconomics).

In the case where government purchases are relevant, what was derived above is
not exactly right. In order to account for it, start with the new aggregate resource
constraint and log-linearize it:
C∗ G∗
Yt = Ct + Gt ⇔ Yt − Y ∗ = Ct − C ∗ + Gt − G∗ ⇔ Ỹt = C̃t + G̃t
Y∗ Y∗
Denote C ∗ /Y ∗ as sC the consumption share (obviously G∗ /Y ∗ is equivalent to
(1 − sC )), and you get:
1 
C̃t =

Ỹt − (1 − sC )G̃t
sC
which you can plug in the log-linearized version of the Euler equation:
1  1    1 
i˜t − Et [πt+1 ]

Ỹt − (1 − sC )G̃t = Et Ỹt+1 − (1 − sC ) Et G̃t+1 −
  
sC sC σ
 sC 
i˜t − Et [πt+1 ] + (1 − sC ) G̃t − Et G̃t+1
  
⇔ Ỹt = Et Ỹt+1 −
 
σ
This equation shows that while permanent government expenditures shock have
no effect on demand, transitory ones will affect demand positively.

91
Labor market equilibrium and the supply curve

The labor market equilibrium is given by the conjunction of labor supply and
labor demand:
η
χHt = λt (Wt /Pt ) and Wt /Pt = Zt /µ
Using the fact that λ = Ct−σ = Yt−σ = (Zt Ht )−σ , we can write:

η 1 η 1 1 1−σ
χHt = · Zt Ct−σ ⇔ Ht = · Zt (Zt Ht )−σ ⇔ Ht = · Ztη+σ
µ χµ χµ
We can use this equilibrium condition to get the aggregate supply of the economy:

1 1+η
Yt = Zt Ht = · Ztη+σ
χµ
This equation makes it obvious what factors can affect the supply curve. They
are: labor preferences ( χ), markup (µ), both IES of labor and consumption (η and
σ) as well as most importantly the level of technology Zt . Dynamically, equation
yields the following log-linearized equation:
1+η
Ỹt = Z̃t
η+σ

Now consider the case with government purchases. In order to come up with an
equation simply, we use the first equation above and log-linearize before replacing
η
consumption by output: χHt = 1µ · Zt Ct−σ ⇔ η H̃t = Z̃t − σC̃t . Now plugging
in the log-linearized version of both the production function and the aggregate
resource constraint, we get:
σ 
η H̃t = Z̃t − σC̃t ⇔ η(Ỹt − Z̃t ) = Z̃t −

Ỹt − (1 − sC )G̃t
sC
ηsC + σ σ(1 − sC )
⇔ Ỹt = (1 + η) Z̃t + G̃t
sC sC
(1 + η)sC σ(1 − sC )
⇔ Ỹt = Z̃t + G̃t
ηsC + σ ηsC + σ
and we see that the supply function is a positive function of government purchase,
regardless of the type of shock.

92
Money market

We now have two equations describing the economy, a demand equation (NKIS)
and a supply equation. However, the NKIS lacks an important element: the price
level and nominal interest rates are not determined by anything. This issue can
be solved in two ways:

• Introducing money in the economy via a money market. Money demand

would be set ad hoc, even though you could model it in households’ UMPs,
as in Walsh; money supply would be set exogenously as a stochastic process
(the same type as technology Z).
• Introducing an interest rate setting authority (i.e. a central bank) following
a rule for interest rates.

Since for now, a Taylor does not make too much sense, let’s focus on the money
market solution. We’ll derive and use the Taylor rule in the section about the
three-equations New-Keynesian model.

As we said earlier, money demand is defined ad hoc by:

Mt γ
= Yt · (1 + it )−ν
Pt
Note that in this notation, the interest rate has a negative power, while in Basu’s
notation you’d only have the ν. The two notation are equivalent since Basu’s ν < 0,
however, I prefer to have it negative directly in the equation. Log-linearizing
money demand gives us:
ln(Mt ) − ln(Pt ) = γ ln(Yt ) − ν ln(1 + it )
M̃t − P̃t = γỸt − νi˜t
The money demand is therefore downward-sloping in the (i, M
P ) space.

The money supply is taken as a completely exogenous stochastic process. It is

written as:
M̃t+1 = ρ M · M̃t + εt+1
M

which is a vertical line in the (i, M) space.

The money market equilibrium can also be described as a curve (the equivalent
of the LM curve) in the (i,Y ) space (where it will be useful for the next part):

93
 • If ν = 0 and γ = 1, then as interest rate goes up, nothing happens in the
money market: the LM curve is vertical in (i,Y ). This means that the interest
rate does not change following shocks:

M̃t − P̃t = Ỹt

• If ν > 0, then as output goes up, the interest also goes up: the LM curve
is upward sloping. You will see in the results section that this will tend to
have a stabilizing effect in the economy.

6.1.2 Results

Technology shocks: Z

In this model, a shock to technology will have the same real effects overall as in
the RBC model without capital. However we need to consider what happens to
prices in addition to real variables.

Recall that a productivity shock has two major consequences: it is a positive

wealth shock for households and it decreases marginal cost (you can produce
more with the same amount of labor). This means that in the labor market, labor
supply shifts to the left and labor demand to the right: real wages rise and hours
worked are ambiguous (as with the RBC model, parametrization solves this issue).

Let’s now turn to the aggregate demand and supply curves. You can clearly see
from the NKIS equation that demand is not directly affected by a technology
shock. On the supply side, what happens is also obvious since output is positively
affected by technology. Therefore output increases (as the vertical supply curve
shifts right), which puts negative pressures on inflation: prices go down.

Since output increases, consumption increases as well. Moreover, if the shock

is only transitory, it will bring the real interest rate down. This is the case since
consumption smoothing requires more savings and without capital, the interest
must go down to leave the market in equilibrium. In a permanent shock, nothing
would happen to the real interest rate.

Y ↑; C ↑; W/P ↑; H?; r ↓ or ∼; π ↓

94
Monetary shocks: M

Since this model is so close to an actual RBC model without capital, a money
supply shock should have no real effects.

Indeed, if you look at the firms’ optimization problem, output is not a function of
the money supply: it must not move. From the money demand perspective, this
means that prices should move. Indeed, recall that M̃t − P̃t = Ỹt so that if Ỹt = 0,
then we must have M̃t = P̃t . This also happens with a positive ν. We would have
M̃t = P̃t − νi˜t
implying that either prices move up (π ↑) or the nominal interest rate goes down
(i˜ ↓). Both dynamics have a positive effect on aggregate demand following the
NKIS equation. This shifts the NKIS up, which creates inflationary pressures, but
no movement to the output, thus no effect on consumption, no effect on the labor
market.
Y ∼; C ∼; W/P ∼; H ∼; r ∼; π ↑

Government shocks: G

Consider now the model with government expenditures. We saw in the previous
sections that transitory and permanent shocks might have different effects on
demand, so we will see transitory shocks first, then permanent ones.

As we have seen, a transitory shock in government expenditures has a positive

effect on aggregate demand, and a positive effect on aggregate supply. This should
have an overall positive effect on output, and an ambiguous effect on prices. Let’s
turn to a more detailed analysis to clear things out. In the labor market, only
labor supply should shift right because of the PIH. This increases hours worked,
leaving real wages intact since labor demand is horizontal. Since production
has increased, the marginal cost has also increased, thus prices increase (fixed
markup). The real interest rate go up because of the PIH.
Y ↑; C ↓; H ↑; W/P ∼; r ↑; π?

Now let’s focus on permanent government shocks. This time aggregate demand
does not move and aggregate supply shifts right. This raises output and yields

95
deflationary pressure. In the labor market, all directions form the transitory are
kept the same although labor supply shifts less (no need for extra smoothing).
Hours worked go up, real wages stay the same. Real interest should not move
since it is a permanent shock.

Y ↑; C ↓; H ↑; W/P ∼; r ∼; π ↓

Markup shocks: µ

A positive markup shock will show up directly in the supply curve, shifting it
left. It may be counter-intuitive for some that while the profit margin goes up,
production goes down, but in fact it follows from the fact that when costs don’t
move, increasing the margin means increasing prices, leading to lower demand
and thus lower production. This shift causes inflationary pressures.

The lower production (lower consumption) is affecting the labor supply which
shifts right (negative wealth effect) while labor demand shifts down (recall W/P =
Z/µ). For output to go down, it must be that the shift in supply in lower in
magnitude than the shift in demand since hours must go down (technology has
not moved). Therefore hours worked decrease, and real wages decrease. As
consumption go down, PIH suggests that real interest rates should go up.

Y ↓; C ↓; H ↓; W/P ↓; r ↑; π ↑

Laziness shocks: χ

As in the previous shock, the labor preferences parameter χ enters directly in

the aggregate supply function and not in the aggregate demand one. A higher
χ implies higher disutility in labor, so that people want to work less. It shifts
aggregate supply left while leaving AD intact, thus reducing output and putting
inflationary pressures. In more detail, we have a shift left of labor supply because
of higher disutility, although it may be slightly less variable as there is also a
negative wealth effect pushing labor supply right. Overall hours decrease and
real wages stay constant. Output decreases, consumption decreases as well, real
interest go up.
Y ↓; C ↓; H ↓; W/P ∼; r ↑; π ↑

96
Patience shocks: β

Contrary to the previous two shocks, a patience shock will only affect demand.
This is hard to see in the log-linearized version of the NKIS but if one looks at the
Euler equation, you clearly see that current consumption is negatively correlated
with the patience parameter β. This would make sense as someone more patience
would value future consumption more than someone impatient, and therefore
would be willing to save more. The issue in this model is that you cannot save
more since there is no capital. The real interest has to go down so that you don’t
save. In the end, nothing has changed, only the real interest has moved down. You
clearly see now that, similar to a RBC model, this model is not demand-driven.

Y ∼; C ∼; H ∼; W/P ∼; r ↓; π ∼

News shocks: Zt+1

A positive news shock will have the same type of effect as the previous shock,
since it comes only through a willingness to save less (you know your future
income is higher), this cannot cause anything to happen because of the lack in
vehicles of savings. This means the real interest rate go up and nothing else
changes.
Y ∼; C ∼; H ∼; W/P ∼; r ↑; π ∼

6.1.3 Interpretation

This stripped down version of an imperfect-competition RBC model without

capital does not have great results. In fact, it can be affected only by real shocks,
which are limited to technology shocks in this model. Any monetary shock will
have zero effect on the economy.

This model was in fact useful only to introduce the sticky prices case, that we will
do in the next section.

97
6.2 NK model without K, sticky prices

6.2.1 Model

Calvo prices

Let’s return to out previous model of pricing, given by the equation:

φt
= 1/µ
Pt
where µ is given by the term θ/(θ − 1). Now suppose that prices are not flexible
enough to change after a shock (we’ll come up with a story for that in a minute).
Then, firms will have to adjust their markup, µt is variable:
φt
= 1/µt ⇔ φ̃t − P̃t = − µ̃t
Pt
Now, recall the labor demand equation in which we had:
Wt
= φt
Zt
Dividing by Pt gives the real marginal cost:
Wt /Pt
= φt /Pt ⇔ W̃t − P̃t − Z̃t = φ̃t − P̃t
Zt
The idea is to use that last equation and all equilibrium conditions to understand
how the markup varies. We can find:

W̃t − P̃t − Z̃t = φ̃t − P̃t ⇔ W̃t − P̃t − Z̃t = − µ̃t

⇔ η H̃t − λ̃t − Z̃t = − µ̃t
⇔ η(Ỹt − Z̃t ) + σC̃t − Z̃t = − µ̃t
⇔ η(Ỹt − Z̃t ) + σỸt − Z̃t = − µ̃t
⇔ (η + σ)Ỹt − (1 + η) Z̃t = − µ̃t
1+η
 
⇔ (η + σ) Ỹt − Z̃t = − µ̃t
η+σ

98
If your memory from the previous model is fresh, the right term in the bracket
should ring a bell: it is the output variation in the flexible prices case! We denote
f
it Ỹt for flexible. Now, what is the difference between the actual output and its
flexible value? The output gap, denoted X̃t . Since it is caused by nominal rigidities
(prices), this output gap was not present in the previous model and as we can
see from the equation above, the output gap is important in determining what
happens to the markup.

Now, back to what would cause a difference between the desired price level and
the actual price level: we need a story that forces firms to keep their prices the
same for longer than they’d want to. Calvo (1983) came up with the idea that, at
every period, a fraction of firms (ω), chosen randomly, are stuck with the price
that they set in the previous period, even if the optimal price is now different. If
ω = 0, we’re back to the flexible prices model. Since firms know that they might
not be selected in the next periods to change their prices, the pricing function
will be very different. In particular, more than the just the markup and/or the
marginal cost, a firm needs to work with its expectations of future prices (if prices
are known to be higher in the future, let’s change them now). In fact, firms are
kind of smoothing their prices as households smooth their consumption. In the
end, after long and tedious calculations (not detailed here), the pricing function is
given by:

f (σ + η)(1 − ω)(1 − βω)

πt = β Et [πt+1 ] + κ(Ỹt − Ỹt ) where κ =
ω
This equation links inflation to the output gap (or unemployment), this is why it
is called the New-Keynesian Phillips Curve (NKPC). The NKPC curve is in this
model equivalent to aggregate supply in the previous one, but this curve slopes
up (compared to the vertical AS is the previous flexible prices case).

Rewriting the NKIS

Since the NKPC is written in terms of inflation and output gap, we might as well
rewrite the NKIS in terms of both variables as well. Doing that is very simple,

99
 h i
f f
substract Ỹt and Et Ỹt+1 :

 1 
i˜t − Et [πt+1 ] − Ỹt − Et Ỹt+1
h i  h i
f f f f
Ỹt − Ỹt − Et Ỹt+1 = Et

Ỹt+1 −
σ
 1 
i˜t − Et [πt+1 ] − Ỹt
h i 
f f
⇔ X̃t − Et Ỹt+1 = Et

X̃t+1 −
σ
 1 
i˜t − Et [πt+1 ] + Et Ỹt+1 − Ỹt
  h i 
f f
⇔ X̃t = Et X̃t+1 −

σ
 1 
i˜t − Et [πt+1 ] + ut

⇔ X̃t = Et X̃t+1 −

σ
The interest rate will be defined ad-hoc as we did in the previous model by an
exogenous money supply and demand.

6.2.2 Results

In this section, let’s use the exact same shocks as in the flexible prices so as to
see more clearly what happens to the output gap and solve the shocks more
intuitively. Since overall, both types of ad hoc money demand yield the same kind
of responses, we assume that the money demand is:

Mt /Pt = Yt

Technology shocks: Z

Let’s recall what happened in the flexible prices case. We had that flexible output
went up due to a positive supply shock. However in the sticky prices case, we
have that M̃t − P̃t = Ỹt , while M̃t = 0 and prices are sticky P̃t cannot allow for Ỹt
f f
to follow Ỹt . This means that in the short run, Ỹt > Ỹt : there is a negative output
gap.

A negative output gap will decrease inflation, and increase the short-run markup.
The increase in markup will lower labor demand (still above its steady-state)
so that hours decrease (which is logic since output increases less than flexible
output while technology is the same). Consumption follows output and thus
increase. Inflation will slowly push Yt back to the flexible output level. Output has

100
 f
a hump-shaped (it catches up with Yt , then follows it while it goes down), thus
consumption has a hump shape as well: the interest rate must jump up (people
want to dissave on impact), then go down.

Monetary shocks: M

Following a monetary shock, we have seen that the flexible output does not
move at all. In the sticky prices case, we have that M̃t = Ỹt since P̃t ≈ 0 (a
useful simplification, even if the real situation is not exactly zero). Therefore,
output must go up following a monetary shock. This creates a positive output gap,
which in turn decreases markup, raising labor demand, hours worked and wages.
Inflation then kicks in and slowly pushes aggregate supply up to counteract the
shift up of aggregate demand.

Government shocks: G

Analyzing this shock in terms of aggregate demand and supply might be slightly
tricky. In fact, it happens that depending on how you shift the curves you could
find either type of output gap. In place of that reasoning, alter the NKIS to include
the output gap and government expenditures:
 sC 
i˜t − Et [πt+1 ] + (1 − sC ) G̃t − Et G̃t+1 + ut
  
X̃t = Et X̃t+1 −
 
σ
then you see that a transitive shock to G will yield a positive output gap (demand
shifts more than long-run supply). Then the analysis is the same way, a positive
output gap means higher inflation and a lower markup. A lower markup means
a higher labor demand curve, increasing hours more than simply the shift in
labor supply. Since government expenditures grow by the same amount as in
the flexible output case and output grows more, it must be that consumption is
also moving up. Higher inflation will slowly brings variables back to their new
(higher) steady-state.

101
Markup shocks: µ

Laziness shocks: χ

Patience shocks: β

News shocks: Zt+1

6.3 Three-equations NK model

In the previous model, we assumed that the real money supply was somehow
linked to the interest rate and output by an ad-hoc relation, completely exogenous
when ν = 0, or endogenous. However, this relation, while simplifying a lot of
computations, is not very realistic as to how the control of money demand. In
particular, it does not take into account the role of central banks. This section uses
the same type analysis that we used before, but instead of using a money market
defined ad hoc, we model the relationship between interest rates and output via a
monetary policy. This new monetary policy will effectively replace the LM curve
as the new money equilibrium curve.

6.3.1 Taylor Rule

As we previously saw, in order for a monetary policy to replace the LM curve it

needs to link interest rates with output. Another major goal of a monetary policy
would be to stabilize the economy. Stabilizing in this context would mean to
allow the economy to be more flexible, to ”ignore” nominal rigidities. Therefore it
seems logical to start with looking at interest rates in the flexible prices case.

102
Natural rate of interest

Take the Euler equation in the flexible prices case:

i 1 
i˜t − Et [πt+1 ]
h 
f f
Ỹt = Et Ỹt+1 −
σ
Replacing the rates by the real interest rate (Fisher equation), we get:
 h i 
f f
rt = σ Et Ỹt+1 − Ỹt

Since it represents a benchmark for the real interest rate, we call it the natural
real interest rate, and denote it rtn . We can plug it in the actual NKIS curve (with
sticky prices) to get:
 1
X̃t = Et X̃t+1 − rt − rtn


σ
This equation shows an interesting concept: the monetary authority can control
the output gap by setting the real interest rate as close to r n as possible. Wicksell
in particular suggested to design the following policy rule:

it = rtn + Et [πt+1 ]

The issue with this policy rule is that because of inderteminacy of the equation
system, the rule would allow for sunspot equilibria.

Taylor principle

We need to fight inflation more than one-for-one:

it = δπ πt

where δπ > 1.

103
6.3.2 Role of the monetary policy

6.3.3 Results

Interest rate shock

Technology shock

6.3.4 Interpretation

Now, think of the aggregate demand in the economy depending on both goods
(consumption) and the money market equilibrium. In this model the supply does
f
not change, it is the NKPC: πt = β Et [πt+1 ] + κ(Ỹt − Ỹt ). On the demand side
however, we now have to use two equations, one summing up the goods market
equilibrium, another one summing up the money market equilibrium.

6.3.5 Demand block

Goods market equilibrium

The goods market equilibrium

h is determined
i by the Euler equation for consump-
tion, namely Ct = Et βCt+1 (1 + it ) Pt+1 . If we log-linearize it, we get:
−σ −σ Pt

−σC̃t = −σ Et C̃t+1 + i˜t − Et [πt+1 ]

 

 1 
C̃t = Et C̃t+1 − i˜t − Et [πt+1 ]
 
σ
Because there is no capital, consumption must equal output and we can replace
consumption growth by output growth.
f
Finally, substracting the growth of output under flexible prices Ỹt to each output
variable, you get the following condition:
 1 
X̃t = Et X̃t+1 − i˜t − Et [πt+1 ] + ut
 
σ

104
 h i
f f
where ut = Et Ỹt+1 − Ỹt+1 . This equation is called the NKIS curve in reference to
it being the analog of the IS curve we have studied in undergrad.

Money Market equilibrium

On to the money market equilibrium, we need to specify an equation for the

behavior of interest rates with regards to output in the money market. Typically,
we specify two types of equations for that purpose: the Taylor rule and the usual
quantity theory of money equation.

The Taylor rule specifies a rule for central banks behavior as:

i˜t = δπ πt + δ X X̃t

This rule implies that central banks choose the nominal interest rate (the money
supply) by reacting to inflation and the output gap at the current period. Note
that this is only an assumption on the central bank behavior and not some result
of solving for a welfare problem of how to set interest rates.

The second rule comes from the QTM and is written as

yt = mt − pt

This gives a vertical relationship between interest rates and output as output
depends only on money supply growth and inflation!

6.3.6 Supply block

The supply equation is not different from the simple NK model we’ve seen in the
previous section. It links inflation and the output gap, it’s the NKPC:

πt = β Et [πt+1 ] + κ X̃t

Since the NKPC is the last equation of this model, let’s go over the whole model
again and analyze its structure. First of all, we can notice that one part relies only
on assumptions, hence we might ask ourselves if the assumptions are correct

105
and what they imply for the model. The next section is going to discuss issues
related to the Taylor rule. Moreover, we can see that the new model has no state
(backward-looking) variable as capital is not included in the model. Again, a
following section will try and implement capital in the NK framework.

6.3.7 Taylor rule: a discussion on the optimal monetary pol-

icy

We define the natural interest rate as the interest rate that would prevail if prices
were flexible (as a counter factual proxy for output at its full potential). Under
flexible prices we have:
 1
˜
h i
f f
X̃t = Et X̃t+1 − (it − Et [πt ]) + Et Ỹt+1 − Ỹt

σ
where X̃ = 0 for all periods as the output gap does not exist. Hence,
i 1
Ỹt = Et Ỹt+1 − (i˜t − Et [πt ])
h
f f
σ
Our definition of the interest rate is therefore:
 h i 
f f
r̃t = σ Et Ỹt+1 − Ỹt
n

that we can plug back in our Euler equation:

 1
X̃t = Et X̃t+1 − (r̃t − r̃tn )

σ

This equation means in words that the current output gap is determined by
expectations of future output gaps (because r n is also a function of output growth
under flexible prices). A good monetary policy’s goal is to achieve the lowest
output gap possible at all times. From this equation, this goal can be sustained by
simply keeping the real interest rate as close as possible to the natural interest
rate. In order to do that, a simple rule would be to set nominal interest rate at:
it = rtn + Et [πt+1 ]
This rule is deemed to not satisfy Taylor’s principle as a rise in inflation expec-
tations is met by a raise of nominal interest rates in the same value. This leads

106
to an unchanged real interest rate: the real economy is unchanged while agents
seem to expect an acceleration, the economy accelerates and the central bank is
behind.

In general, using Taylor’s rule, sufficient conditions for stability imply putting
a weight δπ > 1 (i.e. fighting inflation aggressively or punishing agents for
expecting excess inflation).

6.3.8 IRFs for NK models

Interest Rate Shock

Consider a one-time negative shock of a quarter of a percent to the nominal

interest rate.

From the goods market equilibrium condition, we can see that actual output will
increase instantly, while the flexible prices output does not change. Therefore
the output gap is also increasing, leading to an increase in the inflation rate at
the same period. In the following period, because this model has no inertia, the
interest rate goes back to normal and all variables go back to the steady state: no
persistence. The following graph shows the effect of this shock clearly:

107
In conclusion, we can see that money shocks have no real effects, have no lasting
effects (only on impact) and no hump-shaped responses (as in our VAR estimates):
they describe the reality pretty poorly as is. Nevertheless, we can add elements
to the model so that we have better effects. For example, we could imagine that
interest rates take time to adjust (not hard to believe considering actual behavior
of the FED), we call this phenomenon interest rate smoothing. The effects are of
course in the same directions but we see a longer persistence in these effects.

108
Technology shock

Now considering a shock to technology, we should see an increase in output

with flexible prices and a smaller increase in current output. Therefore we should
also see a decrease in output gap, leading to a decrease in inflation and hence a
reaction of the central bank to reduce nominal interest rates. Hours worked will
go down as the current output increase is not enough to offset the increase in
technology. Real wages should go up.

109
With interest rate smoothing, we get a different story. Indeed, because the interest
rate cannot adjust instantly, it is above the optimal interest rate for some time,
causing a smaller increase in current output (and a bigger decrease in output gap).
This will lead to a lower inflation. Hours worked will decrease more and real
wage will increase less.

110
In conclusion, it seems that while output gaps still exist after a technology shock,
the dynamic neo-keynesian model damps it with an active monetary rule. With
interest rate smoothing, output gaps are higher (since optimal policy takes longer
time to be set). However, this interest rate smoothing appears to create persistent
effects and hump-shaped responses, although reducing welfare.

111
6.3.9 Conclusions

This section has shown that the form of the monetary policy rule has important
implications on the dynamics of the economy. In fact, the NK model with exoge-
nous money had no difference in reactions compared to flexible prices. However,
a Taylor-rule kind of policy has proven to give different responses, especially
with added elements like interest rate smoothing. But this Taylor rule is again an
assumption, moreover an assumption that sets the monetary rule at its optimal.
How can we make this behaviour depart from optimality?

• We could model a ZLB.

• We could model the typical way in which interest rates are smoothed.
• We can add other types of sticky prices (wages, …).

6.4 NK model with investment

This section will give insights as to whether the absence of I and K in our previous
model had any impact on the results we found. In order to look at this relationship,
we’ll analyze the imperfect competition model, adding Calvo frictions in prices.
This will yield interesting results, close to what we derived with variable markups
but in this model, target markups will not change.

6.4.1 Model

All agents in the economy (consumers, investors and government) demand a

composite good Yt with elasticity of substitution σ. The demand is:
∫ 1 σ
 σ−1
σ−1
Yt = yit di
σ

which implies the following demand for each firm i (as derived in the Rotemberg-
Woodford model):   −σ
pit
yit = Yt
Pt

112
The aggregate production function displays increasing returns to scale (and we’ll
assume no profit at the steady state, as we did in the Rotemberg-Woodford model),
hence:
Yt = Ktα (Zt Ht )1−α − Φ
which gives the following log-linearized production function:

Ỹt = µ(1 − sH )K̃t + µsH ( Z̃t + H̃t )

Y ∗ +Φ
where µ = Y∗ is the degree of returns to scale. In the factor markets, we have:

Yt + Φ Yt + Φ
Rt = φt α ; Wt = φt (1 − α)
Kt Ht

Adding Calvo prices, we get the following NKPC curve (as derived during the TA
session):
πt = β Et [πt+1 ] + λ φ̃t
This time φ̃t takes a different form because of imperfect competition. From the
labor demand FOC, we have:
Y∗
W̃t = φ̃t + Ỹt − H̃t
Y∗ + Φ
Y∗
⇔ φ̃t = W̃t + H̃t − ∗ Ỹt
Y +Φ
Y∗
⇔ φ̃t = W̃t + H̃t − Ỹt + Ỹt − Ỹt
Y∗ + Φ
Φ
⇔ φ̃t = s̃tH + Ỹt
Y∗ + Φ

We introduce two types of monetary policies: one exogenous, the other is a

Taylor-rule with interest rate smoothing.

Exog. Money : M̃t − P̃t = γỸt − ν(r̃t + Et [πt+1 ])

Taylor Rule : i˜t = δi i˜t−1 + δπ πt + δ X X̃t + εt

Consumers’ behavior is unchanged from our usual assumptions in all but one
way: they can invest in capital as well as risk-less bonds. This gives a purpose to

113
having two Euler equation:

Euler eq. for bonds: λ̃t = βr ∗ · r̃t+1 + Et λ̃t+1

 

Euler eq. for capital: λ̃t = βR∗ · Et R̃t+1 + Et λ̃t+1

   

which means that by the arbitrage condition, it must be that

R∗  r∗ + δ
r ∗ · r̃t+1 = R∗ · Et R̃t+1 ⇔ r̃t+1 = ∗ · Et R̃t+1 = · Et R̃t+1
    
r r ∗

This last equation is the condition for equilibrium in the assets market. Assuming
that Et R̃t+1 = R̃t+1 can simplify greatly the graphical analysis adding two


interpretations to the model: there’s no uncertainty about interest rates, all

shocks on interest rates are anticipated. From both factor demand equations we
can write:
Rt α Ht
= ⇔ R̃t = W̃t + H̃t − K̃t
Wt 1 − α Kt
and hence we can write:
r∗ + δ
r̃t = [W̃t + H̃t − K̃t ]
r∗
This equation is the NRR curve. We can note that this implies that r̃t is increasing
in Y .

Graphically, the NRR curve

and the LM curve (exoge-
nous money) gives the fol-
lowing relationship, where
NRR is the curve in this
model, compared to NK the
classical NKIS curve from
previous models.

114
6.4.2 IRFs for NKK models

Money supply shocks

As can be made clear by the previous schematic graph, an increase in the money
supply, or a shift to the right of the LM curve, will create an increase in both real
interest rates and output. This will in turn cause a decrease in markups, leading
to a

Technology and Government expenditures shocks

115