Advanced Macroeconomics, 2024-2025

Intertemporal choices over infinite horizon

Etienne Wasmer

NYUAD

February 9, 2025

1/53
Moving towards infinity (2) Ramsey - simplified

▶ Intertemporal choices of capital accumulation and

consumption
▶ Culmination of neo-classical theory of growth (later used to
study business cycles)
▶ Solow + Euler equation.
▶ Transitory dynamics exist and can be quantified.
▶ Require no-Ponzi condition, otherwise easy to generate
instable, unsustainable dynamics.

2/53
Ramsey - simplified

▶ Ramsey (1928), very early, perhaps too early, rediscovered bY

David Cass and Tjalling Koopmans in the early 1960’s.
▶ Agents (dynasty ) live forever. Discrete time.
▶ They consume ct per period.
▶ They have an income denoted by yt coming from the
production side (it will refer later on to the income from labor
or human capital).
▶ Discount the future at rate β < 1 (discount factor).
▶ In the original Ramsey paper, a dynasty grows at some rate n.
Would introduce complications ignored here. Constant
population.

3/53
Notation: utility

▶ Intertemporal utility function

U(c0 , c1 , ..., ct , ...) = u(c0 ) + βu(c1 ) + ... + β t u(ct ) + ...
▶ u ′ > 0, u ′′ < 0
▶ More compact notation:
∞
X
U(c0 , c1 , ..., ct , ...) = β t u(ct )
t=0

4/53
Notation: utility, continued

▶ We will use u(ct ) = ct1−θ /(1 − θ)

u”(ct )ct
▶ R=− is the coefficient of aversion to risk and
u ′ (ct )
intertemporal fluctuations
▶ Reminder: the higher R the lower the desire to substitute
consumption.
▶ here R = θ

5/53
Debt/savings in an infinite horizon

▶ Denote by at the net asset of an agent.

▶ So, at+1 = at (1 + rt ) + yt − ct
▶ If at negative: debt.

6/53
Transversality condition

▶ First assume constant discount R = 1 + r .

▶ We impose that
at R −t → 0
when t → ∞: the non-Ponzi game condition.
▶ Avoids explosive debt.
▶ To get it: write a0 as a function of at+1 = at R + yt − ct
iteratively from 1 to some terminal period T + 1.
▶ How much debt can one afford...
T
−1
X yi − c i
a0 + R = aT +1 R −(T +1)
Ri
i=0

▶ if T is finite, final reimbursement. If infinity, equivalent is the

last term to vanish asymptotically.

7/53
Transversality condition

▶ Now, under time varying discount rates: transversality

becomes:
at (R1 R2 ...Rt )−1 → 0
when t → ∞ Comes from composed debts at time t to
infinity.
▶ Same as last slide, if again constant R.
▶ To get it: write again a0 as a function of at+1 = at Rt + yt − ct
iteratively from 1 to t+1. Not done yet but on slide 16.

8/53
Maximization of the agent

▶ At time t, the agent takes at as given, and chooses ct and

therefore, at+1 .
▶ Complicates a bit by assuming time-varying returns
Rt = 1 + rt .
▶ Program:

Max U(c0 , c1 , ..., ct , ...) = u(c0 )+βu(c1 )+...+β t u(ct )+...

ct ,at+1 ,t≥0

subject to at+1 = at Rt + yt − ct

9/53
Maximization of the agent

▶ Simple intuition, mathematical treatment simplified.

▶ Introduce a Lagrangian
+∞
X
L= β t u(ct ) − λt [at+1 − at (1 + rt ) − yt + ct ]
t=0

▶ Treat now ct and at+1 as independent variables:

10/53
Maximization of the agent
▶ Max Lagrangian
+∞
X
L= β t u(ct ) − λt [at+1 − at (1 + rt ) − yt + ct ]
t=0

▶ Derivative with respect to ct :

β t u ′ (ct ) = λt

which says that λt is the marginal utility to consume in t

periods
▶ Derivative with respect to at+1 :

λt+1 (1 + rt+1 ) = λt

The multiplier decreases as the rate of returns on financial

assets.
11/53
Maximization of the agent

▶ Combine
β t u ′ (ct ) = λt
and
λt+1 (1 + rt+1 ) = λt
▶ Leads to:
β t u ′ (ct ) = β t+1 u ′ (ct+1 )(1 + rt+1 )
or the Euler equation:

u ′ (ct ) = βu ′ (ct+1 )Rt+1

12/53
Maximization of the agent: an intuition

▶ Ignore for the moment the Lagrangian approach. Just

consider the following steps.
▶ First see that ct+1 function of ct at a given at+2 . To see this,
consider that next period one will have

ct+1 + at+2 = at+1 Rt+1 + yt+1

▶ Now use:
at+1 = at Rt + yt − ct
▶ Implies:

ct+1 + at+2 = (at Rt + yt − ct )Rt+1 + yt+1

13/53
Maximization of the agent

▶ From

ct+1 + at+2 = (at Rt + yt − ct )Rt+1 + yt+1

one has
∂ct+1
= −Rt+1
∂ct
for a given at (predetermined) and at+2 (affected only later,
by ct+1 ) and income profile.
▶ Key: if I consume more today, I need to reduce my
consumption tomorrow by Rt+1 if I want to reach the same
net position at+2

14/53
Maximization of the agent
▶ Using
∂ct+1
= −Rt+1
∂ct
derive with respect to ct in the infinite sum:

Max ... + β t u(ct ) + β t+1 u(ct+1 ) + ...

ct

▶ This FOC leads to :

β t u ′ (ct ) + β t+1 (−Rt+1 )u ′ (ct+1 )

▶ Or after dividing by β t , simply Euler with no uncertainty

u ′ (ct ) = βRt+1 u ′ (ct+1 )

▶ The ways of deriving the Euler equations are diverse and can
also be derived from a recursive formulation. See Sargent and
Ljunqvist (Fourth edition).
15/53
Maximization of the agent

▶ It can also be derived from a different Lagrangian

▶ Let’s compose the period-budget constraints.
▶ Start from a0 R0 = −y0 + c0 + a1 , but a1 can be obtained by
▶ a1 R1 = −y
 1 + c1 + a2 so that 
1 −y1 + c1 + a2
▶ a0 = −y0 + c0 +
R0 R1
▶ and recursively,

T  
X −yt + ct aT +1
a0 = +
t=0
R0 R1 ...Rt R0 R1 ...RT

16/53
Maximization of the agent

▶ Now use the non-Ponzi condition to take the limit to infinity,

and we get the intertemporal budget constraint:
+∞  
X −yt + ct
a0 =
R0 R1 ...Rt
t=0

▶ Therefore the Lagrangian is

+∞ +∞ 
"  #
X
t
X −yt + ct
L= β u(ct ) + λ a0 −
R0 R1 ...Rt
t=0 t=0

17/53
Maximization of the agent

+∞ +∞ 
"  #
X
t
X −yt + ct
L= β u(ct ) + λ a0 −
R0 R1 ...Rt
t=0 t=0

▶ Derivative with respect to ct is zero implying:

1
β t u ′ (ct ) = λ
R0 R1 ...Rt
and for the next period, derivative with respect to ct+1 :
1
β t+1 u ′ (ct+1 ) = λ
R0 R1 ...Rt Rt+1
▶ Take the ratio and no surprise,

u ′ (ct ) = βu ′ (ct+1 )Rt+1

18/53
NB: Euler with uncertainty - class 1

u ′ (ct+1 )
βEt Rt+1 =1
u ′ (ct )

19/53
In the special case of the iso-elastic utility
▶ Uses the derivative of utility to get:
−θ
βRt+1 ct+1 = ct−θ

▶ Log linearize:
ct+1 − ct log (βR)
=
ct θ
▶ In continuous time:
c˙t log (βR)
=
ct θ
▶ The higher the desire to smooth (corresponding to a higher
θ), the lower the chosen growth of consumption.
▶ This superposes to the effect of conflicting forces between
preferences for the present β and market conditions about the
future R.
▶ What if βR = 1?
20/53
And with constant returns

▶ Assume now a constant rt and thus Rt :

▶ Leads (unproved, several steps) to a simple consumption rule
similar to that of two periods :
∞
" #
X
c0 = µ a0 R + R −t yt
t=0

▶ where µ = 1 − (βR)1/θ R −1 ? Let’s prove it (use geometric

series)
▶ Interpretation: Current consumption is a fraction of total
wealth (sum of non-human wealth and human wealth). Yearly
µ around 5% with β = 0.95, R = 1.05, θ = 3

21/53
This is true at any time period say t0

▶ Consumption rule :
∞
" #
X
−(t−t0 )
ct0 = µ at0 R + R yt
t=t0

22/53
Application: effect of income shocks

▶ Effect of a contemporaneous and temporary income shock on

yt :
∂ct
=µ
∂yt
▶ Effect of a differed and temporary income shock on yt+T :

∂ct
= µR −T < µ
∂yt+T

▶ Effect of a permanent income shock on all yt ′ = y , t ′ ≥ t:

 
∂ct 1 1+r 1
=µ =µ =µ 1+ >µ
∂y 1 − 1/R r r

23/53
Production side

▶ We now want to endogeneize rt and Rt : connect the Euler

equation to capital accumulation and vice-versa.
▶ Capital Kt

24/53
A preview: the modified Golden rule

▶ With no technical progress, optimal capital stock for agents

leads to:
r ∗ = f ′ (k ∗ ) = σ + n + δ
where β = 1/(1 + σ) the ’psychological’ rate of interest of
agents.
▶ While in Solow, the Golden rule is (with g = 0):

r ∗ = f ′ (k ∗ ) = n + δ
▶ Called Modified Golden Rule: now includes the preference for
the present of agents.
▶ Which raises the returns to capital
▶ That is, reduce accumulation of capital
▶ Because agents prefer consumption today to future (steady)
state-consumption.

25/53
Production side

▶ Production Yt = F (Kt , At Nt ) constant returns to scale (Cobb

Douglas with share α for capital
▶ Per effective unit of labor variables: lower case:
yt = F (Kt , At Nt )/(At Nt ) = F (kt , 1) = f (kt )
▶ For a few slides: no population growth, no technology growth
(n = 0, g = 0).

26/53
Production side, continued

▶ As usual the returns to capital are

f ′ (kt ) = rt

and is also the financial income.

▶ And the wage is what is left

wt = f (kt ) − kt f ′ (kt )

where wt is the human income.

27/53
Production side, continued

▶ However: agents (dynasties) receive a wage AND returns on

capital so the relevant income is

yt = f (kt ) = wt + kt f ′ (kt )

is the some of human income and financial income.

28/53
Combine demand of the consumer and production

▶ Assets are simply the capital stock (assumed liquid, full resale
value, no depreciation):

at = kt

▶ Dynamic accumulation of capital:

kt+1 − kt = f (kt ) − ct

▶ Equivalent to a relation between consumption and savings:

ct = f (kt ) − (kt+1 − kt )

▶ Preferences of consumers:

u ′ (ct ) = β(1 + rt+1 )u ′ (ct+1 )

29/53
Dynamic system

▶ So, joint system in (kt , ct ) connecting ct and kt via

rt = f ′ (kt ).
▶ Transversality condition adapted due to time-varying interest
rate:
β t u ′ (ct )
lim kt+1 = 0
t→∞ u ′ (c0 )

▶ The present value (in utility terms) of capital must converge to

zero in infinity.
▶ This eliminates explosive paths in capital by consuming
little/nothing to over-accumulate capital to infinity.
▶ Guess: what would happen to this constraint if consumption
eventually brought no further utility?

30/53
Steady-state

▶ Is there a steady-state? Assume it exists, that is

ct = ct+1 = c ∗ ; rt = rt+1 = r ∗ ; and kt = kt+1 = k ∗ .
▶ Euler equation

u ′ (ct ) = β(1 + rt+1 )u ′ (ct+1 )

becomes:
1 = β(1 + r ∗ )

31/53
Steady-state

▶ From it,
f ′ (k ∗ ) = r = 1/β − 1
1
▶ We had a notation for β = where σ is a psychological
1+σ
discount factor (and 1 + σ was a MRS).
▶ So r ∗ = σ: the interest rate equals the psychological discount
factor, and
1 + r ∗ = MRS
Everything reconnects! When is consumption maximized? Is
this is a Golden rule ? (when g = n = 0: efficient
consumption?).

32/53
Steady-state with population growth

▶ In Ramsey, it is assumed that a dynasty has a growing size at

rate n.
▶ Then, the consumption per period is automatically reduced
unless income or savings grow.
▶ One can demonstrate that the modified Golden rule will
become
1 + r ∗ = 1 + f ′ (k ∗ ) = (1 + σ)(1 + n)
or
r∗ ≈ σ + n > n

33/53
Intuition for why the formula contains n when it is
positive?
▶ Because now,

at+1 = (1 + n)−1 [at Rt+1 + yt − ct ]

so each period; the accumulated savings are reduced by

(1 + n)−1 or the equivalent rate of return on savings is

Rt+1 /(1 + n) = (1 + rt+1 )/(1 + n)

▶ So the model leads in the steady-state to

(1 + r ∗ )/(1 + n) = (1 + σ)
▶ And then
r ∗ = f ′ (k ∗ ) = σ + n
(first order approximation)
▶ Except, as in most textbooks, if utility discount also includes
population growth (σ becomes σ − n). Not our choice here. 34/53
Steady-state with population growth

▶ NB: exact relation if time period goes to zero

r∗ = σ + n > n

▶ The interest rate has to reflect the growth of population (to

induce agents to save for the future generation) and the rate
of preferences for the present (to compensate for the sacrifice
of consumption).
▶ The more impatient the agents, the higher the rate of interest
that is the lower the stock of capital needed.

35/53
Summary: steady-state consumption and capital

1. The stock of capital no longer evolves, so, in the absence of

depreciation:
f (k ∗ ) = c ∗
2. Capital stock determined by

r ∗ = f ′ (k ∗ ) = σ

▶ If n > 0 then
r ∗ = f ′ (k ∗ ) = σ + n

36/53
Steady-state consumption and capital

▶ If we had depreciation (and no technical progress):

r ∗ = f ′ (k ∗ ) = σ + n + δ

▶ While in Solow, the Golden rule is:

r ∗ = f ′ (k ∗ ) = n + δ
▶ Called Modified Golden Rule: now includes the preference for
the present of agents.
▶ Which raises the returns to capital
▶ That is, reduce accumulation of capital
▶ Because agents prefer consumption today to future (steady)
state-consumption.

37/53
Exercise

▶ Write the problem with capital depreciation

▶ Write the Lagrangian.
▶ Calculate the steady-state return on capital
▶ Calculate the steady-state consumption
▶ Calculate the maximal steady-state consumption

38/53
Steady-state consumption and capital
▶ Additional subtleties with technical progress g
▶ It enters in the modified Golden rule but as θg , that is, as in
the Golden rule only if θ = 1 (Cobb-Douglas utility).
▶ In that case:

r ∗ = f ′ (k ∗ ) = σ + n + δ + θg

as opposed to Solow’s rule

r ∗ = f ′ (k ∗ ) = n + δ + g
Higher g leads to less capital per efficient unit if θ > 0.
Harder to wait for better technology if consumption
smoothing motive (θ > 0)
▶ Anyway... The take away here is the race
▶ between capital accumulation and technical progress (good for
consumption)
▶ and population growth and preferences for the present (bad for
consumption). 39/53
More general method using optimization tools
▶ So far: method of optimization by agents leading to Euler
equation using informally an argument of two consecutive
periods
▶ One was the two periods model (classes 1 and 2)
▶ One was the use of
∂ct+1
∂ct
holding assets constant. Intuitive but not satisfactory since
asset could evolve.
▶ Macroeconomics has developed many optimization tools in
infinite horizon.
▶ Lagrangian
▶ Today: continuous time version of it called Hamiltonian
(denoted by H).
▶ Often lead to an equivalence: recursive systems,
Jacobi-Bellman equations.
▶ Start: what is the difference between physics and economics?
40/53
More general method using optimization tools
▶ Will be useful for consumer theory as well as firm investment
theory, labor market analysis, finance theory.
▶ Suppose that you want to maximize some objective function
Z ∞
U(t, kt , zt )dt
0

subject to an accumulation constraint on kt such that

k˙t = g (t, zt , kt )

▶ and an initial condition k0

▶ and a terminal condition limt→∞ kt bt ≥ k̄ where bt is an
arbitrary function that has a finite limite in infinity, such as
bt = (1 + r )−t for instance.
▶ zt is a control variable and kt is a state variable.

41/53
More general method using optimization tools

▶ The Infinite Horizon Maximum Principle

▶ Assume U and g to be continuously differentiable.
▶ Introduce the Hamiltonian and a new unknown µt as

H(t, kt , zt , µt ) = U(t, kt , zt ) + µt g (t, zt , kt )

▶ µt is a multiplier and is called the co-state of the state

variable kt .

42/53
More general method using optimization tools

▶ The Infinite Horizon Maximum Principle (continued)

▶ The solution for kt and zt must satisfies:

∂H(t, kt , zt , µt )
=0
∂z
∂H(t, kt , zt , µt )
= −µ˙t
∂k
▶ It exists and is unique under some assumptions, typically of
(strict) concavity with respect to kt of H (actually its max
with respect to zt ).
▶ That’s (almost) all you need to know in continuous time
macro. A good and exhaustive presentation to it is Chapter 7 (pp 227-285) of Introduction to
Modern Economic Growth, MIT Press, by Daron Acemoglu.

43/53
Discounting in continuous time

1
▶ Start from β = if time period last 1.
1+σ
▶ If periods last an arbitrary small unit of time dt,

1
β= ≈ 1 − σdt
1 + σdt
▶ Continuous time: dt → 0. We usually weight utility by e −σt
▶ Why? When we go from t to t + dt, the discount is

e −σdt ≈ 1 − σdt

▶ Exponential discounting is the continuous time equivalent to

geometric series/

44/53
Application to our problem of optimal growth

▶ Suppose that you want to maximize

Z ∞
e −σt u(ct )dt
0

▶ with an accumulation constraint on kt that

k˙t = f (kt ) − ct − δkt

▶ and an initial condition k0

45/53
Application to our problem of optimal growth

▶ The first order condition on ct leads to

µt = u ′ (ct )e −σt

which states that the multiplier is the discounted value of

consumption for the consumer.
▶ And, for capital,

∂H(t, kt , ct , µt )
= µ(t) f ′ (kt ) − δ = −µ˙t
 
∂k

46/53
Application to our problem of optimal growth

▶ Combine them, to get:

−u ′′ (ct )  ′ 
ċt = f (kt ) − σ − δ
u ′ (ct )

k˙t = f (kt ) − ct − δkt

47/53
Gets back to our dynamic equations

▶ With iso-elastic utility function:

ċt f ′ (kt ) − σ − δ
=
ct θ

k˙t = f (kt ) − ct − δkt

▶ We are now equipped to study the dynamics of capital

accumulation and consumption.

48/53
Dynamics, when σ > α, δ = n = 0

49/53
Dynamics, when σ > α, δ = n = 0

50/53
Definition of a saddle-path

▶ For each possible initial value of k0 , there is a unique level of

consumption that allows the economy to converge to the
steady-state.
▶ If agents were to chose another level of consumption, one
would diverge.
▶ Hence converge on this saddle-path, the only one consistent
with knowledge of the model by the agents, and of the
common-knowledge rationality of agents.

51/53
Characterization of the steady-state consumption and
capital with a Cobb-Douglas and n, δ = 0
1. Consumption
(k ∗ )α = c ∗

2. Capital
α (k ∗ )α−1 = σ

▶ Combined:
σ ∗
c∗ = k
α
Philippe Weil’s presentation, pre-print of Macroeconomic Theory: A Primer

▶ NB: if n > 0:
σ+n ∗
k c∗ =
α
▶ NB2: if δ > 0: c ∗ no longer a simpler function of k ∗
52/53
Conclusion of the neo-classical growth model

1. Consumption per person can be stable, as well as capital

stock per person
2. Or per efficient unit of technical progress.
3. Capitalism may be stable after all, but requires smart agents.
4. Note however that mistakes can be corrected in real time:
ct = c ∗ (kt ).
5. With heterogeneous agents: may be more difficult to
coordinate, especially with heterogeneous beliefs.

53/53