14.

452 Economic Growth: Lecture 13, Stochastic Growth

Daron Acemoglu

MIT

December 10, 2013.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 1 / 52

 Stochastic Growth Models Stochastic Growth Models

Stochastic Growth Models

Brock and Mirman (1972): generalization of neoclassical growth and
starting point of Real Business Cycle models
Baseline neoclassical growth: complete markets, households and …rms
can trade using any Arrow-Debreu commodity.
Complete markets: full set of contingent claims traded competitively.
Implies that individuals can fully insure against idiosyncratic risks.
Source of interesting uncertainty thus is aggregate shocks.
Bewley (1970s and the 1980s): households cannot use contingent
claims and can only trade in riskless bonds.
Explicitly prevent risk-sharing and thus “incomplete markets”.
Stochastic stream of labor income: can only achieve smoothing via
“self-insurance”.
Does not admit a representative household; trading in contingent
claims not only su¢ cient, but also necessary for representative
household assumption with uncertainty.
Key for study of questions related to risk, income ‡uctuations and
policy.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 2 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model I

With competitive and complete markets, the First and Second

Welfare Theorems so equilibrium growth path is identical to the
optimal growth path.
But analysis is more involved and introduces new concepts.
Economy as baseline neoclassical growth model, but production
technology now given by

Y (t ) = F (K (t ) , L (t ) , z (t )) , (1)

z (t )=stochastic aggregate productivity term

Suppose z (t ) follows a monotone Markov chain (as de…ned in
Assumption 16.6) with values in the set Z fz1 , ..., zN g.
Many applications assume aggregate production function takes the
form Y (t ) = F (K (t ) , z (t ) L (t )).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 3 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model II

Assume that the production function F satis…es usual assumptions
and de…ne
Y (t )
y (t )
L (t )
f (k (t ) , z (t )) ,
Fraction δ of the existing capital stock depreciates at each date.
Suppose z1 , ..., zN are arranged in ascending order and that j > j 0
implies f (k, zj ) > f (k, zj 0 ) for all k 2 R+ .
Thus higher values of the stochastic shock z correspond to greater
productivity at all capital-labor ratios.
Representative household with instantaneous utility function u (c )
that satis…es the standard assumptions.
Supplies one unit of labor inelastically, so K (t ) and k (t ) can be used
interchangeably (and no reason to distinguish C (t ) from c (t )).
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 4 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model III

Consumption and saving decisions at time t are made after observing

z (t ).
Sequence version of the expected utility maximization problem of a
social planner:
∞
max E0 ∑ βt u (c (t )) (2)
t =0

subject to

k (t + 1) = f (k (t ) , z (t )) + (1 δ ) k (t ) c (t ) and k (t ) 0,
(3)
with given k (0) > 0.
To characterize the optimal growth path using the sequence problem:
de…ne feasible plans, mappings k̃ [z t ] and c̃ [z t ] with
zt (z (0) , ..., z (t )).
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 5 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model IV

Instead look at the recursive version:
V (k, z ) = max u f (k, z ) + (1 δ) k k 0 + βE V k 0 ,
k 0 2[0,f (k ,z )+(1 δ)k ]
(4)
Proposition In the stochastic optimal growth problem described above,
the value function V (k, z ) is uniquely de…ned, strictly
increasing in both of its arguments, strictly concave in k and
di¤erentiable in k > 0. Moreover, there exists a uniquely
de…ned policy function π (k, z ) such that the capital stock
at date t + 1 is given by k (t + 1) = π (k (t ) , z (t )).
Proof: verifying that Assumptions 16.1-16.6 from the previous
chapter are satis…ed and apply Theorems.
To do this, …rst de…ne k̄ such that k̄ = f (k̄, zN ) + (1 δ) k̄, and
show that starting with k (0) 2 (0, k̄ ), the capital-labor ratio will
always remain within the compact set (0, k̄ ).
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 6 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model V

Proposition In the stochastic optimal growth problem described above,
the policy function for next period’s capital stock, π (k, z ),
is strictly increasing in both of its arguments.
Proof:
By assumption u is di¤erentiable and from the Proposition above V is
di¤erentiable in k.
By the same argument as before, k 2 (0, k̄ ); thus we are in the interior
of the domain of the objective function.
Thus, the value function V is di¤erentiable in its …rst argument and
u 0 f (k, z ) + (1 δ) k k0 βE V 0 k 0 , z 0 j z = 0,
Proposition above: V is strictly concave in k. Thus this can hold when
k or z increases only if k 0 also increases.
For example, an increase in k reduces the …rst-term (because u is
strictly concave), hence an increase in k 0 is necessary to increase the
…rst term and to reduce the second term (by the concavity of V ).
Argument for increase in z is similar.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 7 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model VI

De…ne the policy function for consumption as

π c (k, z ) f (k, z ) + (1 δ) k π (k, z ) ,

where π (k, z ) is the optimal policy function for next date’s capital
stock determined in Proposition above.
Using this notation, the stochastic Euler equation can be written as

u 0 (π c (k, z )) = βE f 0 π (k, z ) , z 0 + (1
δ) u 0 π c π (k, z ) , z 0 j
(5)
A di¤erent way of expressing this equation makes it both simpler and
more intuitive:

u 0 (c (t )) = βEt p (t + 1) u 0 (c (t + 1)) , (6)

p (t + 1) is the stochastic marginal product of capital (including

undepreciated capital) at date t + 1.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 8 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model VII

Also useful for comparison with the competitive equilibrium because

p (t + 1) corresponds to the stochastic (date t + 1) dividends paid
out by one unit of capital invested at time t.
Proposition above characterizes form of the value function and policy
functions, but:.
1 Not an analog of the “Turnpike Theorem”: does not characterize the
long-run behavior of the neoclassical growth model under uncertainty.
2 Qualitative results about the value and the policy functions, but no
comparative static results.
Stochastic law of motion of the capital-labor ratio:

k (t + 1) = π (k (t ) , z (t )) , (7)

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 9 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model VIII

De…nes a general Markov process, since before the realization of z (t ),

k (t + 1) is a random variable, with its law of motion governed by the
last period’s value of k (t ) and the realization of z (t ).
If z (t ) has a non-degenerate distribution, k (t ) does not typically
converge to a single value.
But may hope that it will converge to an invariant limiting
distribution.
Markov process (7): starting with any k (0), converges to a unique
invariant limiting distribution.
I.e., when we look at su¢ ciently faraway horizons, the distribution of
k should be independent of k (0).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 10 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

The Brock-Mirman Model IX

Moreover, the average value of k (t ) in invariant limiting distribution

will be the same as the time average of fk (t )gTt=0 as T ! ∞
(stochastic process for the capital stock is “ergodic”).
A “steady-state” equilibrium now corresponds not to speci…c values
but to invariant limiting distributions.
If z (t ) takes values within a su¢ ciently small set, this limiting
invariant distribution would hover around some particular values
(“quasi-steady-state” values)
But in general the range of the limiting distribution could be quite
wide.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 11 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

Example: Brock-Mirman with Closed-form Solution I

Suppose u (c ) = log c, F (K , L, z ) = zK α L1 α, and δ = 1.

Again z follows a Markov chain over the set Z fz1 , ..., zN g, with
transition probabilities denoted by qjj 0 .
Let k K /L. The stochastic Euler equation (5):
" #
1 αz 0 π (k, z )α 1
= βE 0 z , (8)
zk α π (k, z ) z π (k, z )α π (π (k, z ) , z 0 )

Relatively simple functional equation in a single function π ( , ).

Here “guessing and verifying” is handy. Conjecture that

π (k, z ) = B0 + B1 zk α .

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 12 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

Example: Brock-Mirman with Closed-form Solution II

Substituting this guess into (8):
1
(9)
(1 B1 ) zk α B0
" #
αz 0 (B0 + B1 zk α )α 1
= βE 0 z .
z (B0 + B1 zk α )α B0 B1 z 0 (B0 + B1 zk α )α
This equation cannot be satis…ed for any B0 6= 0.
Thus imposing B0 = 0 and writing out the expectation explicitly with
z = zj 0 , this expression becomes
1 N αzj (B1 zj 0 k α )α 1
(1 B1 ) zj 0 k α
= β ∑ jj zj (B1 z 0 k α )α B1 zj (B1 z 0 k α )α .
q 0
j =1 j j

Simplifying each term within the summation:

N
1 α
(1 B1 ) zj 0 k α
= β ∑ qjj 0
B1 ( 1 B1 ) z j 0 k α
.
j =1
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 13 / 52
 The Brock-Mirman Model Optimal Growth under Uncertainty

Example: Brock-Mirman with Closed-form Solution III

Now taking zj 0 and k out of the summation and using the fact that,
by de…nition, ∑Nj =1 qjj 0 = 1, we can cancel the remaining terms and
obtain
B1 = αβ,
Thus irrespective of the exact Markov chain for z, the optimal policy
rule is
π (k, z ) = αβzk α .
Identical to deterministic case, with z there corresponding to a
non-stochastic productivity term.
Thus stochastic elements have not changed the form of the optimal
policy function.
Same result applies when z follows a general Markov process rather
than a Markov chain.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 14 / 52

 The Brock-Mirman Model Optimal Growth under Uncertainty

Example: Brock-Mirman with Closed-form Solution IV

Here can fully analyze the stochastic behavior of the capital-labor

ratio and output per capita.
Stochastic behavior of the capital-labor ratio in this economy is
identical to that of the overlapping generations model
But just one of the few instances of the neoclassical growth model
that admit closed-form solutions.
In particular, if the depreciation rate of the capital stock δ is not
equal to 1, the neoclassical growth model under uncertainty does not
admit an explicit form characterization.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 15 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty I

Environment identical to that in the previous section, z an aggregate

productivity shock a¤ecting all production units and households.
Arrow-Debreu commodities de…ned so that goods indexed by di¤erent
realizations of the history z t correspond to di¤erent commodities.
Thus economy with a countable in…nity of commodities.
Second Welfare Theorem applies and implies that the optimal growth
path characterized in the previous section can be decentralized as a
competitive equilibrium
Moreover, since we are focusing on an economy with a representative
household, this allocation is a competitive equilibrium without any
redistribution of endowments.
Justi…es the frequent focus on social planner’s problems in analyses of
stochastic growth models in the literature.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 16 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty II

But explicit characterization of competitive equilibria shows the

equivalence, and introduces ideas related to pricing of contingent
claims.
Complete markets: in principle, any commodity, including any
contingent claim, can be traded competitively.
In practice no need to specify or trade all of these commodities; a
subset su¢ cient to provide all necessary trading opportunities.
Will also show what subsets are typically su¢ cient.
Preferences and technology as in previous model: economy admits
representative household and production side can be represented by a
representative …rm.
Household maximize the objective function given by (2) subject to the
lifetime budget constraint (written from the viewpoint of time t = 0).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 17 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty III

No loss of generality in considering the viewpoint of time t = 0

relative to formulating with sequential trading constraints.
Z t =set of all possible histories of the stochastic variable z t up to
date t .
Z ∞ =set of in…nite histories.
z t 2 Z ∞ =a possible history of length t.
p0 [z t ]=price of the unique …nal good at time t in terms of the …nal
good of date 0 following a history z t ,
c [z t ] and w0 [z t ] similarly de…ned.
Household’s lifetime budget constraint:
∞ ∞
∑ ∑ p0 z t c z t ∑ ∑ w0 z t + k (0) . (10)
t =0 z t 2Z ∞ t =0 z t 2Z ∞

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 18 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty IV

No expectations:
Complete markets: all trades at t = 0 at price vector for all
Arrow-Debreu commodities.
Household buys claims to di¤erent “contingent” consumption bundles;
i.e. conditioned on z t .
Left-hand=total expenditure taking the prices of all possible claims as
given.
Right-hand side=labor earnings and value of initial capital stock per
capita.
Right-hand side of (10) could also include pro…ts accruing to the
individuals, but constant returns and competitive markets implies that
equilibrium pro…ts will be equal to 0.
Objective function at time t = 0:
∞
∑ βt ∑ q zt j z0 u c zt , (11)
t =0 z t 2Z ∞

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 19 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty V

q z t j z 0 =probability at time 0 that the history z t will be realized at
time t.
Sequence problem of maximizing (11) subject to (10). Assuming
interior solution, …rst-order conditions: is
βt q z t j z 0 u 0 c z t = λp0 z t (12)
for all t and all z t .
λ is the Lagrange multiplier on (10) corresponding to the marginal
utility of income at date t = 0
Combining two di¤erent date t histories z t and ẑ t :
u 0 (c [ẑ t ]) p0 [ẑ t ] /q ẑ t j z 0
= ,
u 0 (c [z t ]) p0 [z t ] /q [z t j z 0 ]
Right-hand side=relative price of consumption claims conditional on
histories z t and ẑ t
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 20 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty VI

Combining for histories z t and z t +1 such that z t +1 = (z t , z (t + 1)):
βu 0 c z t +1 p0 z t +1 /q z t +1 j z 0
= ,
u 0 (c [z t ]) p0 [z t ] /q [z t j z 0 ]
Right-hand side=contingent interest rate between date t and t + 1
conditional on z t (and contingent on the realization of z t +1 ).
To characterize equilibrium need prices p0 [z t ], from the pro…t
maximization problem of …rm.
R0 [z t ]=price of one unit of capital after the state z t
K e [z t ] and L [z t ] =capital and labor employment levels of the
representative …rm after history z t .
Value of the …rm:
∞
p0 [z t ] (F (K e [z t ] , L [z t ] , z (t )) + (1 δ) K e [z t ])
∑ βt ∑ R0 [z t ] K e [z t ] w0 [z t ] L [z t ]
t =0 z t 2Z ∞

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 21 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty VII

Pro…t maximization implies:

∂F (K e [z t ] , L [z t ] , z (t ))
p0 z t + (1 δ ) = R0 z t
∂K e
∂F (K e [z t ] , L [z t ] , z (t ))
p0 z t = w0 z t .
∂L
Using constant returns to scale:

(13)
t 0 e t
p0 z f k z , z (t ) + (1 δ) = R0 z t
p0 z t f k e z t , z (t ) k e z t f 0 k e z t , z (t ) = w0 z t

Relation between prices and marginal productivity of factors.

But (13) also stating that R0 [z t ] is equal to the value of dividends
paid out by a unit of capital inclusive of undepreciated capital.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 22 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty VIII

Alternative, equivalent, way of formulating competitive equilibrium
and writing (13) is to assume that capital goods are rented.
Labor market clearing:
L z t = 1 for all z t . (14)
Production after history z t is f (k e [z t ] , z (t )) + (1 δ) k e [z t ],
divided between consumption c [z t ] and savings s [z t ].
Capital used at time t + 1 (after history z t +1 ) must be equal to s [z t ] .
Market clearing for capital implies that for any z t +1 = (z t , z (t + 1)),
k e z t +1 = s z t , (15)
Capital market clearing condition:
c zt + s zt f s zt 1
, z (t ) + (1 δ) s z t 1
(16)
for any z t +1 = (z t , z (t + 1)).
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 23 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty IX

Capital market clearing condition also implies no arbitrage condition

linking R0 z t +1 to p0 [z t ].
Consider the following riskless arbitrage:
Buy one unit of the …nal good after z t to be used as capital at time
t + 1 and simultaneously sell claims on capital goods for each
z t +1 = (z t , z (t + 1)).
No risk, since unit of …nal good bought after history z t will cover the
obligation to pay capital good after any z t +1 = (z t , z (t + 1)).
Implies the no arbitrage condition

p0 z t = ∑ R0 z t , z (t + 1) . (17)
z (t +1 )2Z

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 24 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty X

Competitive equilibrium: c [z t ] , s [z t ] , k e z t +1 z t 2Z t , and
fp0 [z t ] , R0 [z t ] , w0 [z t ]gz t 2Z t , such that households maximize utility
(i.e., satisfy (12)), …rms maximize pro…ts (i.e., satisfy (13) and (17)),
and labor and capital markets clear (i.e., (14), (15), and (16) are
satis…ed).
Substitute from (13) and (17) into (12) and rearrange:
λp0 z t +1
u0 c z t = ∑ βt q [z t j z 0 ]
f 0 k z t +1 , z (t + 1 ) + (1 δ)
z (t +1 )2Z
(18)
Next using (12) for t + 1:
λp0 z t +1
βu 0 c z t +1 =
β t q [ z t +1 j z 0 ]
λp0 z t +1
= ,
β t q [ z t +1 j z t ] q [ z t j z 0 ]
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 25 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XI

Second line uses the law of iterated expectations,

q z t +1 j z 0 q z t +1 j z t q z t j z 0 .
Substituting into (18), we obtain

u0 c z t
f 0 k z t +1 , z (t + 1 )
= β ∑ q z t +1 j z t u 0 c z t +1
+ (1 δ )
z (t +1 )2Z

= βE u 0 c z t +1 f 0 k z t +1 , z (t + 1 ) + (1 δ) j z t ,

Identical to (6).
Proposition In the above-described economy, optimal and competitive
growth path coincide.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 26 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XII

Equilibrium problem in its equivalent form with sequential trading

rather than all trades taking place at the initial date t = 0.
Write the budget constraint of the representative household
somewhat di¤erently.
Normalize the price of the …nal good at each date to 1.
a [z t ]s=Basic Arrow securities that pay out only in speci…c states on
nature.
fa [z t ]gz t 2Z t =set of contingent claims that the household has
purchased that will pay a [z t ] units of the …nal good at date t when
history z t is realized.
Price of claim to one unit of a [z t ] at time t 1 after history z t 1
denoted by p̄ z (t ) j z t 1 , where z t = z t 1 , z (t ) .
Amount of these claims purchased by the household is denoted by
a z t 1 , z (t ) .
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 27 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XIII

Thus ‡ow budget constraint of the household:

c zt + ∑ p̄ z (t + 1) j z t a zt 1
, z (t ) w zt + a zt ,
z (t +1 )2Z

w [z t ]=equilibrium wage rate after history z t in terms of …nal goods

dated t.
Let a denote the current asset holdings of the household (realization
of current assets after some z t has been realized).
Then ‡ow budget constraint of the household can be written as

c+ ∑
0
p̄ z 0 j z a0 z 0 j z w + a,
z 2Z

Function p̄ [z 0 j z ]=prices of contingent claims (for next date’s state

z 0 given current state z).
a0 [z 0 j z ]=corresponding asset holdings.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 28 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XIV

V (a, z )=value function of the household.

Choice variables: a0 [z 0 j z ] and consumption today, c [a, z ].
q [z 0 j z ] =probability that next period’s stochastic variable will be
equal to z 0 conditional on today’s value being z.
Then taking the sequence of equilibrium prices p̄ as given, the value
function of the representative household:

u (a + w ∑z 0 2Z p̄ [z 0 j z ] a0 [z 0 j z ])
V (a, z ) = sup .
fa 0 [z 0 jz ]gz 0 2Z + β ∑z 0 2Z q [z 0 j z ] V (a0 [z 0 j z ] , z 0 )
(19)
All Theorems on the value function can again be applied to this value
function.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 29 / 52

 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XV

First-order condition for current consumption:

∂V (a0 [z 0 j z ] , z 0 )
p̄ z 0 j z u 0 (c [a, z ]) = βq z 0 j z
∂a
for any z 0 2 Z .
Capital market clearing:

a 0 z 0 j z = a 0 [z ] ,

Thus in the aggregate the same amount of assets will be present in all
states at the next date.
Thus …rst-order condition for consumption can be alternatively
written as
∂V (a0 [z ] , z 0 )
p̄ z 0 j z u 0 (c [a, z ]) = βq z 0 j z . (20)
∂a
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 30 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XVI

No arbitrage condition implies

∑
0
p̄ z 0 j z R z 0 j z = 1, (21)
z 2Z

where R [z 0 j z ] is the price of capital goods when the current state is

z 0 and last period’s state was z.
Intuition:
Cost of one unit of the …nal good now, 1, has to be equal to return of
carrying it to the next period and selling it as a capital good then.
Summing over all possible states z 0 tomorrow must have total return of
1 to ensure no arbitrage
Combine (20) with the envelope condition

∂V (a, z )
= u 0 (c [a, z ]) ,
∂a
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 31 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XVII

Multiply both sides of (20) by R [z 0 j z ] and sum over all z 0 2 Z to

obtain the …rst-order condition of the household as

u 0 (c [a, z ]) = β ∑
0
q z 0 j z R z 0 j z u 0 c a0 , z 0 .
z 2Z
= βE R z 0 j z u 0 c a0 , z 0 jz .

Market clearing condition for capital, combined with the fact that the
only asset in the economy is capital, implies:

a = k.

Therefore …rst-order condition can be written as

u 0 (c [k, z ]) = βE R z 0 j z u 0 c k 0 , z 0 jz

which is identical to (6).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 32 / 52
 Equilibrium Growth under Uncertainty Equilibrium Growth

Equilibrium Growth under Uncertainty XVIII

Again shows the equivalence between the social planner’s problem

and the competitive equilibrium path.
Social planner’s problem (the optimal growth problem) is considerably
simpler, characterizes the equilibrium path of all the real variables and
various di¤erent prices are also straightforward to obtain from the
Lagrange multiplier.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 33 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models I

Real Business Cycle (RBC): one of the most active research areas in
the 1990s and also one of the most controversial.
Conceptual simplicity and relative success in matching certain
moments of employment, consumption and investment ‡uctuations
vs. the absence of monetary factors and demand shocks.
But exposition of RBC model useful for two purposes:
1 one of the most important applications of the neoclassical growth
model under uncertainty
2 new insights from introduction of labor supply choices into the
neoclassical growth model under uncertainty generates.
Only di¤erence is instantaneous utility function of the representative
household now takes the form

u (C , L) ,

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 34 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models II

u is jointly concave and continuously di¤erentiable in both of its

arguments and strictly increasing in C and strictly decreasing in L.
Also assume that L has to lie in some convex compact set [0, L̄].
Focus on the optimal growth formulation: maximization of
∞
E ∑ βt u (C (t ) , L (t ))
t =0

subject to the ‡ow resource constraint

K (t + 1) F (K (t ) , L (t ) , z (t )) + (1 δ ) K (t ) C (t ) .

z (t ) again represents an aggregate productivity shock following a

monotone Markov chain.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 35 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models III

Social planner’s problem can be written recursively as
u (F (K , L, z ) + (1 δ) K K 0 , L)
V (K , z ) = sup
L 2[0,L̄ ] + βE [V (K 0 , z 0 ) j z ]
K 0 2[0,F (K ,L,z )+(1 δ)K ]
(22)
Proposition The value function V (K , z ) de…ned in (22) is continuous
and strictly concave in K , strictly increasing in K and z, and
di¤erentiable in K > 0. There exist uniquely de…ned policy
functions π k (K , z ) and π l (K , z ) that determine the level of
capital stock chosen for next period and the level of labor
supply as a function of the current capital stock K and the
stochastic variable z.
Assuming an interior solution, relevant prices can be obtained from
the appropriate multipliers and the standard …rst-order conditions
characterize the form of the equilibrium.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 36 / 52
 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models IV

De…ne the policy function for consumption:

π c (K , z ) F K , π l (K , z ) , z + (1 δ) K π k (K , z ) ,

Key …rst order conditions (write π J short for π J (K , z ) , J = c, l, k):

h
uc π c , π l = βE R π k , z 0 uc π c π k , z 0 , π l π k , z 0

w ( K , z ) uc π c , π l = ul π c , π l .

where

R (K , z ) = Fk (K , z ) + (1 δ)
w (K , z ) = Fl (K , z )

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 37 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models V

First condition in (23) is essentially identical to (5), whereas the

second is a static condition determining the level of equilibrium (or
optimal) labor supply.
Second condition does not feature expectations: conditional on the
current value K and the current z.
Analysis of macroeconomic ‡uctuations: period in which z is low.
If no o¤setting change in labor supply, “recession”.
Under standard assumptions, w (K , z ) and labor supply decline: low
employment and output.
If Markov process for z exhibits persistence, persistent ‡uctuations.
Provided F (K , L, z ) is such that low output is associated with low
marginal product of capital, expectation of future low output will
typically reduce savings and thus future levels of capital stock
This e¤ect depends also on form of utility function (consumption
smoothing and income and substitution e¤ects).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 38 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Application: Real Business Cycle Models VI

Thus model may generate some of the major qualitative features of

macroeconomic ‡uctuations.
RBC literature argues it generates the major quantitative features
such as correlations between output, investment, and employment.
Debate on whether:
1 the model did indeed match these moments in the data;
2 these were the right empirical objects to look at; and
3 focusing on exogenous changes in aggregate productivity sidestep why
there are shocks.
RBC debate is not as active today as it was in the 1990s, but not a
complete agreement.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 39 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Example: RBC model with closed-form solution I

u (C , L) = log C γL, F (K , L, z ) = zK α L1 α, and δ = 1.

z follows a monotone Markov chain over the set Z fz1 , ..., zN g,
with transition probabilities denoted by qjj 0 .
Conjecture that
π k (K , z ) = BzK α L1 α
.
Then with these functional forms, the stochastic Euler equation for
consumption (23) implies
" (1 α )
#
1 αz 0 BzK α L1 α
(L0 )1 α
= βE z ,
(1 B ) zK α L1 α
(1 B ) z 0 (BzK α L1 α )α (L0 )1 α

where L0 denotes next period’s labor supply.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 40 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Example: RBC model with closed-form solution II

Canceling constants within the expectations and taking terms that do
not involve z 0 out of the expectations:
1 h i
α 1 α 1
= βE α BzK L z ,
zK α L1 α
which yields
B = αβ.
Resulting policy function for the capital stock is therefore

π k (K , z ) = αβzK α L1 α
,

which is identical to that in Example before.

Next, considering the …rst-order condition for labor:
(1 α) zK α L α
= γ.
(1 B ) zK α L1 α

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 41 / 52

 Equilibrium Growth under Uncertainty Application: Real Business Cycle Models

Example: RBC model with closed-form solution III

The resulting policy function for labor as

(1 α)
π l (K , z ) = ,
γ (1 αβ)

Labor supply is constant: with the preferences as speci…ed here, the

income and the substitution e¤ects cancel out, increase in wages
induced by a change in aggregate productivity has no e¤ect on labor
supply.
Same result obtains whenever the utility function takes the form of
U (C , L) = log C + h (L) for some decreasing and concave function h.
Replicates the covariation in output and investment, but does not
generate labor ‡uctuations.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 42 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model I

Economy is populated by a continuum 1 of households and the set of

households is denoted by H.
Each household has preferences given by (2) and supplies labor
inelastically.
Suppose also that the second derivative of this utility function, u 00 ( ),
is increasing.
E¢ ciency units that each household supplies vary over time.
In particular, each household h 2 H has a labor endowment of z h (t )
at time t, where z h (t ) is an independent draw from the set
Z [zmin , zmax ], where 0 < zmin < zmax < ∞.
Labor endowment of each household is identically and independently
distributed with distribution function G (z ) de…ned over [zmin , zmax ].
Production side is the same as in the canonical neoclassical growth
model under certainty.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 43 / 52
 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model II

Only di¤erence is L (t ) is now the sum (integral) of the heterogeneous
labor endowments of all the agents:
Z
L (t ) = z h (t ) dh.
h 2H
Appealing to a law of large numbers type argument, we assume that
L (t ) is constant at each date and we normalize it to 1.
Thus output per capita in the economy can be expressed as
y (t ) = f (k (t )) ,
with k (t ) = K (t ).
No longer any aggregate productivity shock; only uncertainty at the
individual level (i.e., it is idiosyncratic).
Individual households will experience ‡uctuations in their labor
income and consumption, but can imagine a stationary equilibrium in
which aggregates are constant over time.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 44 / 52
 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model III

Focus on such a stationary equilibrium: wage rate w and the gross

rate of return on capital R will be constant .
First take these prices as given and look at the behavior of a typical
household h 2 H
Maximize (2) subject to the ‡ow budget constraint

ah (t + 1) Rah (t ) + wz h (t ) c h (t )

for all t, where ah (t ) is the asset holding of household h 2 H at time

t.
Consumption cannot be negative, so c h (t ) 0.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 45 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model IV

Requirement that individual should satisfy its lifetime budget

constraint in all histories imposes the endogenous borrowing
constraint:
zmin
ah (t )
R 1
b,

for all t.
Maximization problem of household h 2 H recursively:
n h io
V h (a, z ) = sup u Ra + wz a0 + βE V h a0 , z 0 j z .
a 0 2[ b,Ra +wz ]
(24)

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 46 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model V

Proposition The value function V h (a, z ) de…ned in (24) is uniquely
de…ned, continuous and strictly concave in a, strictly
increasing in a and z, and di¤erentiable in
a 2 ( b, Ra + wz ). Moreover, the policy function that
determines next period’s asset holding π (a, z ) is uniquely
de…ned and continuous in a.
Proposition The policy function π (a, z ) derived in Proposition ?? is
strictly increasing in a and z.

Total amount of capital stock in the economy=asset holdings of all

households in the economy, thus in a stationary equilibrium:
Z
k (t + 1) = ah (t ) dh
Zh 2H
= π ah (t ) , z h (t ) dh.
h 2H

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 47 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model VI

Integrates over all households taking their asset holdings and the
realization of their stochastic shock as given.
Both the average of current asset holdings and also the average of
tomorrow’s asset holdings must be equal by the de…nition of a
stationary equilibrium.
Recall policy function a0 = π (a, z ) de…nes a general Markov process:
under fairly weak it will admit a unique invariant distribution.
If not economy could have multiple stationary equilibria or even there
might be problems of non-existence.
Ignore this complication and assume the existence of a unique
invariant distribution, Γ (a), so stationary equilibrium capital-labor
ratio is: Z Z
k = π (a, z ) d Γ (a) dG (z ) ,
which uses the fact that z is distributed identically and independently
across households and over time.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 48 / 52
 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model VII

Turning to the production side:

R= f 0 (k ) + (1 δ )
w = f (k ) k f 0 (k ) .

Recall neoclassical growth model with complete markets and no

uncertainty implies unique steady state in which βR = 1, i.e.,
1
f 0 (k ) = β (1 δ) , (25)

where k refers to the capital-labor ratio of the neoclassical growth

model under certainty.
In Bewley economy this is no longer true.
Implication: dynamic ine¢ ciency possible as in Solow and OLG
models— but for di¤erent reasons.
Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 49 / 52
 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model VIII

Proposition In any stationary equilibrium of the Bewley economy, we

have that the stationary equilibrium capital-labor ratio k is
such that
f 0 (k ) < β 1 (1 δ ) (26)
and
k >k , (27)
where k is the capital-labor ratio of the neoclassical growth
model under certainty.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 50 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model IX

Sketch of proof:
Suppose f 0 (k ) β 1 (1 δ).
Then each household’s expected consumption is strictly increasing.
This implies that average consumption in the population, which is
deterministic, is strictly increasing and would tend to in…nity.
This is not possible since aggregate resources must always be …nite.
This establishes (26).
Given this result, (27) immediately follows from (25) and from the
strict concavity of f ( ).

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 51 / 52

 Growth with Incomplete Markets The Bewley Model

Growth with Incomplete Markets: The Bewley Model X

Interest rate is “depressed” relative to the neoclassical growth model

with certainty because each household has an additional
self-insurance (or precautionary) incentive to save.
These additional savings increase the capital-labor ratio and reduce
the equilibrium interest rate.
Two features, potential shortcomings, are worth noting:
1 Ine¢ ciency from overaccumulation of capital unlikely to be important
for explaining income per capita di¤erences across countries.
model is not interesting because of this but as an illustration of
stationary equilibrium in which aggregates are constant while individual
households have uncertain and ‡uctuating consumption and income
pro…les.
2 Incomplete markets assumption in this model may be extreme.

Daron Acemoglu (MIT) Economic Growth Lecture 13 December 10, 2013. 52 / 52