Notes Consumption Saving 2024
Notes Consumption Saving 2024
Spring 2024
1 Introduction
Macro is about general equilibrium: all agents must behave optimally and all markets must clear.
You have previously learned about decision problems in which prices are taken as given (e.g. the
permanent income hypothesis for consumption). In general equilibrium, the real interest rate is
the intertemporal price of consumption, and adjusts to clear asset markets. We can dress up the
problem to have lots of other markets (labor, investment goods, etc.), but the same general principle
holds.
This note is going to work through simple, two-period, consumption-saving models. I like
to call these “endowment economies,” because there is no production and income is exogenous.
Though simple, there are some powerful insights. In an endowment economy, the real interest
rate adjusts so that agents simply consume their endowment each period – there is no possibility
of moving resources across time, so the real interest rate adjusts to undo all the consumption-
smoothing you’ve previously learned about. As a result, the equilibrium real interest rate can tell
us something about (i) how plentiful the future is expected to be relative to the present and (ii)
how uncertain the future is.
We can also use this simple framework to show how differences at the micro level (i.e. hetero-
geneity) don’t necessarily matter for aggregate dynamics. This justifies the use of representative
agent models. To justify the use of representative agent models, you need complete asset markets
that allow agents to trade at the micro-level.
2 Deterministic Model
There are a continuum of agents populating the unit interval. These agents are indexed by i ∈ [0, 1].
Agents live for two periods, t and t + 1. Agents are endowed with an exogenous and deterministic
stream of income, Yt (i) and Yt+1 (i). Think of the units of income being fruit. Fruit is non-storable.
1
In period t, agents determine how much to consume, Ct (i), and how much to save. Agents have
access to a one-period riskless bond, Bt (i), that trades at price qt and pays out one-for-one in t + 1.
Agents are price-takers.
The budget constraint facing an agent in period t is:
Note that Bt (i) is a stock variable – how much (fruit) one takes from t to t + 1. Bt (i) > 0
means an agent is saving, and Bt (i) < 0 means an agetn is borrowing. In principle, agents could
carry fruit from t + 1 into t + 2, but we disallow this as a terminal condition – agents won’t want
to die with positive asset holdings, and are not permitted to die in debt. Hence, Bt+1 (i) = 0. The
t + 1 budget constraint is:
Agents have a flow utility function given by u(·), where u′ (·) > 0 and u′′ (·) ≤ 0. Agents discount
future utility flows by 0 < β < 1. Lifetime utility is:
The decision problem of an agent is to pick a consumption sequence, Ct (i) and Ct+1 (i), and
bond holdings, Bt (i), to maximize lifetime utility subject to the two budget constraints:
A Lagrangian is:
L = u(Ct (i)) + βu(Ct+1 (i)) + λ1,t (i) (Yt (i) − Ct (i) − qt Bt (i)) + λ2,t (i) (Yt+1 (i) + Bt (i) − Ct+1 (i))
∂L
= 0 ⇐⇒ u′ (Ct (i)) = λ1,t (i)
∂Ct (i)
∂L
= 0 ⇐⇒ βu′ (Ct+1 (i)) = λ2,t (i)
∂Ct+1 (i)
2
∂L
= 0 ⇐⇒ qt λ1,t (i) = λ2,t (i)
∂Bt (i)
L = u(Ct (i))+βu(Ct+1 (i))+λt (i) (Yt (i) − Ct (i) − qt Bt (i))+βλt+1 (i) (Yt+1 (i) + Bt (i) − Ct+1 (i))
Here, instead of λ1,t (i) and λ2,t (i), I just have one Lagrange multiplier indexed by
time, and I discount future constraints by β.1 The FOC will be:
∂L
= 0 ⇐⇒ u′ (Ct (i)) = λt (i)
∂Ct (i)
∂L
= 0 ⇐⇒ u′ (Ct+1 (i)) = λt+1 (i)
∂Ct+1 (i)
∂L
= 0 ⇐⇒ qt λt (i) = βλt+1 (i)
∂Bt (i)
3
in t, and savings brought from t to t + 1 payout principle plus interest, 1 + rt , where rt
is the (net) interest rate (Rt = 1 + rt ) is gross). The budget constraints would be:
Note that Bt (i) is a stock variable – how much (fruit) one takes from t to t + 1.
Bt (i) > 0 means an agent is saving, and Bt (i) < 0 means an agetn is borrowing. In
principle, agents could carry fruit from t+1 into t+2, but we disallow this as a terminal
condition – agents won’t want to die with positive asset holdings, and are not permitted
to die in debt. Hence, Bt+1 (i) = 0. The t + 1 budget constraint is:
With this formulation, the two budget constraints can be combined into one, in-
tertemporal budget constraint:
L = u(Ct (i))+βu(Ct+1 (i))+λ1,t (i) (Yt (i) − Ct (i) − Bt (i))+λ2,t (i) (Yt+1 (i) + (1 + rt )Bt (i) − Ct+1 (i))
∂L
= 0 ⇐⇒ u′ (Ct (i)) = λ1,t (i)
∂Ct (i)
∂L
= 0 ⇐⇒ βu′ (Ct+1 (i)) = λ2,t (i)
∂Ct+1 (i)
∂L
= 0 ⇐⇒ λ1,t (i) = λ2,t (i)(1 + rt )
∂Bt (i)
4
t or t + 1 value of the bond to be one?
The key insight from (4), or alternatively (8), is that all agents will choose the same ratio
of consumption across time because they face a common asset price (qt or 1/Rt ). To see this
concretely, suppose that the utility function is natural log, so the Euler equation becomes:
Ct (i)
1 = β(1 + rt ) (9)
Ct+1 (i)
Alternatively, the (gross) growth rate of consumption between t and t + 1, must be the same
for all agents, and equal to the aggregate growth rate:
This is the market-clearing condition because there is no way, in the aggregate, to transfer
resources across time (the good, fruit, is non-storable, and there is nothing like physical capital).
Individuals could borrow or save, but in aggregate there can be no-borrowing or saving. Integrating
the budget constraints across i, and letting them hold with equality, we get:
Z 1 Z 1 Z 1
Ct (i)di + qt Bt (i)di = Yt (i)di
0 0 0
Z 1 Z 1 Z 1
Ct+1 (i)di = Yt+1 (i)di + Bt (i)di
0 0 0
Using the market-clearing condition for bonds, and defining aggregate consumption and income
as the sums across agents, we have:
C t = Yt (11)
Ct+1 = Yt+1 (12)
In other words, aggregate consumption just equals the aggregate endowment each period.
The competitive equilibrium of this economy is a price, rt , and consumption bundles, Ct (i)
and Ct+1 (i), such that all agents behave optimally and all markets clear. But this is particularly
straightforward to compute and we need not reference individual heterogeneity because the Euler
5
equation implies that agents choose the same growth rate of consumption. So the competitive
equilibrium is characterized by:
1 βu′ (Yt+1 )
= qt = (13)
1 + rt u′ (Yt )
In other words, it is as though there is one agent in the economy – i.e. there exists a rep-
resentative agent. We can determine the aggregate prices and quantities without reference to
household-level heterogeneity. This does not require that all agents are ex-ante identical, nor that
they are even ex-post identical.
To see this concretely, suppose that there are two kinds of agents. Type 1 agents have an
endowment stream of (Yt (1), Yt+1 (1)) = (1, 0), and type 2 agents have an endowment stream of
(Yt (2), Yt+1 (2)) = (0, 1). Suppose that there is a mass, α ∈ [0, 1] of type 1 agents, 1 − α of type 2
agents. Suppose that both types of agents have log preferences. This means that the intertemporal
budget constraints for both types of agents can be written:
Yt+1 (i)
(1 + β)Ct (i) = Yt (i) + (14)
1 + rt
This mean that the consumption and savings function for type 1 agents are:
1
Ct (1) = (15)
1+β
β
Bt (1) = (16)
1+β
Note that both consumption and saving for the type 1 agents are independent of the value of
the real interest rate, rt . This is because the income and substitution effects of rt exactly cancel
out with log preferences. For type 2 agents, we have:
1
Ct (2) = (17)
(1 + β)(1 + rt )
1
Bt (2) = − (18)
(1 + β)(1 + rt )
For type 2 agents, consumption demand is decreasing in rt , and saving supply is increasing in
rt . The aggregate market-clearing condition is the same:
Z 1
Bt (i)di = αBt (1) + (1 − α)Bt (2)
0
Which is:
β 1−α
α − =0
1 + β (1 + β)(1 + rt )
6
Which may be written:
αβ(1 + rt ) − (1 − α) = 0
Or:
1 1−α
1 + rt = (19)
β α
This is the equilibrium (gross) interest rate as a function of exogenous parameters. We can then
solve for the consumption bundles of both types of agents as:
1
Ct (1) = (20)
1+β
1 βα
Ct (2) = (21)
1+β1−α
In other words, the aggregate market-clearing condition is just Ct = Yt (which will also be the
case for t + 1). The aggregate enddowment in t + 1 will be 1 − α. This means that equilibrium real
interest rate, from (19), satisfies:
1 Yt+1
1 + rt = (22)
β Yt
Hopefully you’re catching onto something here. The equilibrium of this endowment economy
with two types of agents is identical to the equilibrium economy where there is one, representative
agent. The representative agent problem with log utility would be:
max ln Ct + β ln Ct+1
Ct ,Ct+1 ,Bt
s.t.
Ct + Bt ≤Yt
Ct ≤Yt+1 + (1 + rt )Bt
The optimality condition for the representative agent is the standard Euler equation:
7
Ct
1 = β(1 + rt ) (23)
Ct+1
Imposing market clearing for the representative agent, we would have Bt = 0, implying Ct = Yt
and hence Ct+1 = Yt+1 , giving:
1 Yt+1
1 + rt = (24)
β Yt
Which is identical to (22). Put differently: even though individual households are neither ex-
ante identical (nor ex-post, because they will, in general, have different consumption bundles), the
aggregate economy behaves as though there is just one household. Basically, what we need for this
result is that agents can freely trade with one another.
u′ (Yt+1 )
1 = β(1 + rt ) (25)
u′ (Yt )
For simplicity, assume log utility, so we have:
Yt
1 = β(1 + rt )
Yt+1
Or:
Yt+1
Yt = (26)
β(1 + rt )
(26) is essentially an “IS curve” in undergraduate parlance (IS: investment equals saving). It
shows how much fruit is demanded in the present as a function of the future future endowment,
Yt+1 , the real interest rate, rt , and a preference parameter governing impatience, β.
8
Figure 1: IS Graph
𝑟𝑟𝑡𝑡
𝐼𝐼𝐼𝐼
𝑌𝑌𝑡𝑡
The IS curve shifts right if Yt+1 is higher (or left if lower). It shifts down (equivalently left) if
β goes up (i.e. if there is an increase in patience).
You graphically compute the equilibrium interest rate by combining this demand relationship
with a supply relationship. Supply in this economy is very straightforward – it’s exogenous (in-
elastic). Call this the Y S curve, given at some exogenous level of the current endowment. This is
shown below:
𝑌𝑌𝑌𝑌
𝑟𝑟𝑡𝑡
𝑟𝑟𝑡𝑡0
𝐼𝐼𝐼𝐼
𝑌𝑌𝑡𝑡
𝑌𝑌𝑡𝑡0
Suppose that there is an increase in the current aggregate endowment, from Yt0 to Yt1 . Let’s
call this a supply shock. We can see how the equilibrium real interest rate will react:
9
Figure 3: Supply Shock
𝑌𝑌𝑌𝑌 𝑌𝑌𝑌𝑌
𝑟𝑟𝑡𝑡
𝑟𝑟𝑡𝑡0
𝑟𝑟𝑡𝑡1
𝐼𝐼𝐼𝐼
𝑌𝑌𝑡𝑡
𝑌𝑌𝑡𝑡0 𝑌𝑌𝑡𝑡1
We see that the equilibrium interest rate must call whenever the current endowment increases.
Let’s stop a second and think about why. Holding the interest rate fixed, we know from our earlier
analysis of consumption (i.e. the PIH), agents would like to increase their saving so as to smooth
out their consumption relative to their income. But, in equilibrium in an endowment economy like
this, it is not possible to increase saving. The real interest rate must fall in an amount sufficient
to make saving more unattractive. In equilibrium, the representative household will just eat the
extra endowment today, even though, for a given interest rate, it would like to smooth.
Now let’s consider a “demand shock.” In particular, suppose the representative household knows
that its future endowment, Yt+1 , is going to be higher. Holding the interest rate fixed, this makes
the household want to consume more in the present (equivalently, save less, or borrow more). In
other words, aggregate demand goes up (the IS curve shifts out). But, in equilibrium the household
cannot save less – it has to just consume its current endowment, which is unchanged. To make the
household satisfied with this, the real interest rate must rise. We can see this in the graph below.
10
Figure 4: Demand Shock
𝑌𝑌𝑌𝑌
𝑟𝑟𝑡𝑡
𝑟𝑟𝑡𝑡1
𝑟𝑟𝑡𝑡0
𝐼𝐼𝐼𝐼
𝐼𝐼𝐼𝐼
𝑌𝑌𝑡𝑡
𝑌𝑌𝑡𝑡0
3 Stochastic Model
Having established some work above, for now let us assume that there is just a representative
agent. The basic consumption-saving problem is the same as before, but the future endowment is
uncertain.
To make life as simple as possible, suppose there are two states of the world in the future: state
1 and state 2 (e.g. high endowment, low endowment). State 1 materializes with probability p, and
state 2 with probability 1 − p. The household want to maximize expected lifetime utility:
11
Note that the expectations operator is taken over expected utility, Et u(Ct+1 ), not utility of
expected consumption. The period t budget constraint is the same as it ever was:
Ct + Bt ≤ Yt (27)
There are two states of the world, and the budget constraint in period t + 1 must hold in both
states of the world (in other words, the budget constraint in t+1 must not only hold in expectation,
it must hold ex-post as well). This means there are two budget constraints in t + 1:
The household’s problem is to pick Ct , Bt , and hence Ct+1 (1) and Ct+1 (2), to maximize expected
utility subject to these budget constraints. Form a current value Lagrangian:
∂L
= 0 ⇐⇒ u′ (Ct ) = λt
∂Ct
∂L
= 0 ⇐⇒ λt = β(1 + rt )(λt+1 (1) + λt+1 (2))
∂Bt
∂L
= 0 ⇐⇒ pu′ (Ct+1 (1)) = λt+1 (1)
∂Ct+1 (1)
∂L
= 0 ⇐⇒ (1 − p)u′ (Ct+1 (2)) = λt+1 (2)
∂Ct+1 (2)
u′ (Ct ) = β(1 + rt ) pu′ (Ct+1 (1)) + (1 − p)u′ (Ct+1 (2)) = β(1 + rt )Et u′ (Ct+1 )
(30)
βu′ (Ct+1 )
1
qt = = Et (31)
1 + rt u′ (Ct )
I can write it this way because I can move 1/u′ (Ct ) and β inside or outside the expectations
operator at will. This is the same Euler equation we had before, just with an expectations operator.
It is a special case of the general point that the price of the bond equals the expected value of the
12
′
product of the stochastic discount factor, βuu′(C t+1 )
(Ct ) , with the payout on the asset (in this case, 1).
Note that, rather than assuming just two future states, we can consider more general probability
distributions over the future endowment. We can write the Lagrangian more succinctly as:
Note that in writing the Lagrangian expectations are over future utility; the future budget
constraint must hold ex-post in any state of the world. The FOC are:
∂L
= 0 ⇐⇒ u′ (Ct ) = λt
∂Ct
∂L
= 0 ⇐⇒ λt = β(1 + rt )λt+1
∂Bt
∂L
= 0 ⇐⇒ Et u′ (Ct+1 ) = λt+1
∂Ct+1
The basic logic of how rt (equivalently, qt ) adjusts to demand and supply is the same as before.
However, with the future endowment uncertain, there is the potential for a different mechanism,
and that is uncertainty. To see this concretely, suppose that we have log utility, that Yt = 1,
and that Et Yt+1 = 1. Suppose now that there is no uncertainty over the future endowment. The
equilibrium real interest rate is easy to compute:
1 Yt+1 1
1 + rt = = (33)
β Yt β
So, for example, if β = 0.98, rt = 0.0204. Now suppose that the future endowment is uncertain.
In particular, suppose there are two states, Yt+1 (1) = 1.5 and Yt+1 (2) = 0.5, each occurring with
probability 1/2. With log utility, the equilibrium real interest rate will be equal to:
1 1 1
1 + rt = ′
β Yt Et u (Yt+1 )
The expected value of future marginal utility is not the marginal utility of expected future con-
sumption/endowment (an application of Jensen’s inequality). Working this out using the numbers
above, we have:
−1 −1
1 1 1 1 1 1 2 1 3 31
1 + rt = × + × = 2 +2 = 2× =
β 2 1.5 2 0.5 β 3 β 8 4β
In other words, the equilibrium real interest rate in this particular example where the future
endowment is uncertainty is 3/4 its value when there is no uncertainty. Why is the equilibrium real
13
rate lower? Because of precautionary saving. When the future is uncertain, the household would
like to save more than when the future is deterministic. But, in equilibrium, the household cannot
save more. Hence, the real interest rate has to be lower to make the household content to just eat
its current endowment.
et is a mean-zero random variable. Suppose it takes on two values: 1/2 and −1/2. In other
words, sometimes the type 1 agent gets more of the endowment and sometimes the type 2 agent
gets more. This means that there are effectively two states of nature in the future facing each agent,
even though there is no aggregate uncertainty. Assume that agents can buy state-contingent bonds
in period t. There are two bonds for both states. B1,t trades at q1,t and pays out when et = 1/2.
B2,t trades at q2,t and pays out one-for-one when et = −1/2.
The period t budget constraints are:
The t + 1 budget constraints facing each type of agent in both states of nature are therefore:
14
1 1
L = u(Ct (1))+ βu(C1,t+1 (1))+ βu(C2,t+1 (1))+λt (1) (Yt − Ct (1) − Bt (1) − q1,t B1,t (1) − q2,t B2,t (1))
2 2
+ βλ1,t+1 (1) (Yt+1 + 1/2 + (1 + rt )Bt (1) + B1,t (1) − C1,t+1 (1)) +
βλ2,t+1 (Yt+1 − 1/2 + (1 + rt )Bt (1) + B2,t (1) − C2,t+1 (1))
The notation here is admittedly poor. I’m using 1 and 2 to denote the two states of the world
(as well as the two agents). The FOC are:
∂L
= 0 ⇐⇒ u′ (Ct (1)) = λt (1)
∂Ct (1)
∂L
= 0 ⇐⇒ λt (1) = (1 + rt )β [λ1,t+1 (1) + λ2,t+1 (1)]
∂Bt (1)
∂L
= 0 ⇐⇒ λt (1)q1,t = βλ1,t+1 (1)
∂B1,t (1)
∂L
= 0 ⇐⇒ λt (1)q2,t = βλ2,t+1 (1)
∂B2,t (1)
∂L 1
= 0 ⇐⇒ u′ (C1,t+1 (1)) = λ1,t+1 (1)
∂C1,t+1 (1) 2
∂L 1
= 0 ⇐⇒ u′ (C2,t+1 (1)) = λ2,t+1 (1)
∂C1,t+1 (1) 2
The first is the standard pricing contingent for the non-state-contingent bond. In the last two,
I have introduced the notation that Dj,t+1 is the payout of each bond, j = 1, 2. D1,t+1 = 1 in state
15
1 and 0 otherwise, and the oppostie for D2,t+1 . The last two conditions look a little funny, but
they are just specialized cases of the general stochastic discount factor pricing condition. The price
you are willing to pay for an asset is just how you value that asset’s payout in relative to how you
value consumption today (i.e. the marginal rate of substitution).
One important thing to note. The FOC for both the type 1 and type 2 agents will take the
identical form – the additive uncertainty in the budget constraints is irrelevant for the FOC. To
make life a little easier, suppose that we have log utility. Since the FOC are the same, and the
agents face the same prices, we have:
Ct (1) Ct (2)
=
C1,t+1 (1) C1,t+1 (2)
Ct (1) Ct (2)
=
C2,t+1 (1) C2,t+1 (2)
In other words, the relative consumption of the two types of agents is the same, across both time
and across states. Suppose that there is a mass 1/2 of each type of agent. The bond market-clearing
conditions are:
1 1
Bt (1) + Bt (2) = 0
2 2
1 1
B1,t (1) + B1,t (2) = 0
2 2
1 1
B2,t (1) + B2,t (2) = 0
2 2
For the state-contingent bonds, this means that, if one type of agent is saving (has a positive
position), the other must be borrowing (has a negative position), i.e. B1,t (1) = −B1,t (2). Let us
now integrate the period t + 1 budget constraints together across agents for each state:
1 1 1 1
C1,t+1 (1) + C1,t+1 (2) = Yt+1 + (1 + rt ) (Bt (1) + Bt (2)) + (B1,t (1) + B1,t (2))
2 2 2 2
1 1 1 1
C2,t+1 (1) + C2,t+1 (2) = Yt+1 + (1 + rt ) (Bt (1) + Bt (2)) + (B2,t (1) + B2,t (2))
2 2 2 2
But with the market-clearing conditions, these just become:
1 1
C1,t+1 (1) + C1,t+1 (2)) = Yt+1 (41)
2 2
1 1
C2,t+1 (1) + C2,t+1 (2)) = Yt+1 (42)
2 2
16
I will skip a few steps, but we get exactly the same thing for the period t budget constraints,
in particular:
1 1
Ct (1) + Ct (2) = Yt (43)
2 2
(41)-(43) simply say that aggregate consumption equals the aggregate endowment in each period
and in each state of nature in the future.
Now, combine the two aggregate resource constraints in the two states of nature for t+1 together
to get:
In other words, consumption for type 2 agents in t + 1 will be identical regardless of which
idiosyncratic state materializes. But if this is true for type 2 agents, then it’s also true for type
1 agents in t + 1. Which then also means that each type of agent has the same consumption in
period t. Which means that the whole problem collapses to a representative agent problem. The
idiosyncratic risk does not matter for aggregate dynamics, because it is functionally insured away
with the state-contingent bonds, B1,t and B2,t . As long as we have a set of complete markets that
cover all idiosyncratic risk, we can sweep that under the rug and justify a representative agent
representation of the problem.
s.t.
Here, Bt−1 (the initial stock of debt) is given (assume zero). The idea is to pick a sequence
of consumption and saving to maximize expected lifetime utility. We can form a current value
Lagrangian:
17
∞
X
L = E0 β t [u(Ct ) + λt (Yt + (1 + rt−1 )Bt−1 − Ct − Bt )]
t=0
Note that I am switching up timing notation here a bit in the formulation of the Lagrangian
relative to the specification of the problem. When time “starts” is arbitrary – I wrote the problem
as starting in t and going forward. I wrote the Lagrangian as starting in period 0 and going forward.
The starting point is completely arbitrary – what matters is correctly keeping track of the past,
the future, and the present. It saves space to start the Lagrangian in period 0 – because then you
can just index time by t instead of t + j.
The FOC are:
∂L
= 0 ⇐⇒ u′ (Ct ) = λt
∂Ct
∂L
= 0 ⇐⇒ λt = βEt λt+1 (1 + rt )
∂Bt
Combining, we get:
βu′ (Ct+1 )
1 = Et (1 + rt ) (47)
u′ (Ct )
This is the same Euler equation as before. It is necessary for an optimal consumption plan.
But it is not sufficient. There will in general be many consumption plans consistent with the Euler
equation. What we need to close things out is a terminal condition. In the two-period model,
the terminal condition was simple – we imposed that the household died with no savings (nor any
debt). Here that is a little more complicated. The transversality condition here is:
The transversality condition says that the expected discounted utility value of any leftover
savings “at the end of time” is zero. λt+T is the current value multiplier (marginal utility of
consumption at t + T , measured in t + T utils). Multiplication by β T discounts it back to the
present. The transversality condition says you place zero value on any leftover saving. This rules
out over-saving (or over-borrowing).
18
s.t.
C + B ′ = Y + (1 + r)B
The value of being in a particular state, V (·), is the maximized value of the current flow
objective function, u(C), plus the discounted value of being in the next state, subject to the budget
constraint. In writing the budget constraint, I have made a slight change in notation for the timing
on bond holdings: B is what you enter the period with, and B ′ is what you choose today that is
available to you tomorrow. V (·) is an unknown function, but assume that it is differentiable.
Re-write the problem by eliminating C and write it in terms of choosing the future state (if
there were no endogenous state variable, the problem would effectively be static):
u Y + (1 + r)B − B ′ + βEV (Y ′ , B ′ )
V (Y, B) = max
′
B
We need a first a first order condition to take care of the max operator. Differentiate with
respect to B ′ and set equal to zero.
∂V (Y ′ , B ′ )
u′ Y + (1 + r)B − B ′ = βE
∂B ′
Now, since we don’t know the function V (·), this doesn’t look so useful. But we can make use
of an envelope condition (Benveniste-Scheinkman) to make progress. Assume that B ′ has been
chosen optimally, so the max operator is taken care of:
∂V (Y, B) dB ′ ∂V (Y ′ , B ′ ) dB ′
= (1 + r)u′ (Y + (1 + r)B − B ′ ) − u′ (Y + (1 + r)B − B ′ ) + βE
∂B dB ∂B ′ dB
∂V (Y ′ , B ′ ) dB ′
∂V (Y, B) ′ ′ ′ ′
= (1 + r)u (Y + (1 + r)B − B ) − u (Y + (1 + r)B − B ) − βE
∂B ∂B ′ dB
But the term in brackets is zero from the first-order condition! So we are left with:
∂V (Y, B)
= (1 + r)u′ (Y + (1 + r)B − B ′ ) = (1 + r)u′ (C)
∂B
Hence:
∂V (Y ′ , B ′ )
= (1 + r)u′ (C ′ )
∂B ′
So, plugging back into the FOC:
19
u′ (C) = β(1 + r)Eu′ (C ′ ) (49)
This is exactly the same FOC we get with the Lagrangian formulation.
βu′ (Yt+1 )
1 = Et (1 + rt )
u′ (Yt )
Since 1+rt is known relative to the expectation operator, we can write the equilibrium condition
as:
′ −1
βu (Yt+1 )
1 + rt = Et (50)
u′ (Yt )
From this, we can intuit how rt will react to more or less persistent changes in income. Suppose
that Yt increases but there is no change in expected future endowments (i.e. a transitory change in
income). u′ (Yt ) will fall a lot, which will push up the stochastic discount factor, which means that
rt must fall. In contrast, suppose that there is a permanent change in income. Yt+1 will go up as
much as Yt , so the stochastic discount factor won’t change, and hence rt won’t change much (or at
all).
What is going on here? rt is adjusting to undo the saving behavior that the household wants to
do. When faced with a transitory increase in income, the household wants to increase its saving.
But in equilibrium it can’t. So the real interest rate must fall to make the household content to
not increase its saving. In contrast, when there is a permanent change in income, the household
doesn’t want to adjust its saving/borrowing behavior – it is fine just eating its extra endowment.
But then the real interest rate doesn’t need to change much (or at all).
20