Macro II

Professor Griffy

UAlbany

Spring 2025
Announcements

I Homework 5 due next Thursday, homework 6 due May 8th.

I Today: Start heterogeneous agent model.
I First: Huggett and Aiyagari.
I Schedule: no class on Thursday (4/10)
I Come to conference instead if you can!
Thinking about Uncertainty in Macroeconomic Models
I Typical assumptions in macroeconomics are a convex
combination of
1. certainty equivalence:

u 0 (c̄i,t ) = E [(1 + rt+1 ) u 0 (c̄i,t+1 ) ] (1)

|{z} | {z }
GE Closer to Linear

2. linearized decision rules:

N
X
((1 + rt+1 )ai,t+1 + wi,t+1 ci,t+1 ai,t+2 ) = 0 (2)
i=1
N
X
((1 + rt+1 ) a Ŝi,t+1 + w (Ŝi,t+1 ) c Ŝi,t+1 a Ŝi,t+2 ) =0
i=1
(3)

I Trick in Krusell-Smith: assume that workers make a linear

prediction about prices in the future.
I i.e., workers use OLS to predict future prices.
Heterogeneous Agent Models
I Workers change their behavior in response to uncertainty.
I First wave of heterogeneous agent models: how do aggregates
change when individual idiosyncratic uncertainty is
uninsurable.
I In other words: when agents must accumulate precautionary
savings to insure against income shocks.
I Key “first wave” papers (no particular order):
I Huggett (1993): Incomplete markets exchange economy with
GE interest rate.
I Imrohoroglu (1989): Individual and aggregate uncertainty with
fixed interest rate.
I Aiyagari (1994): Incomplete markets production economy with
GE interest rate.
I Bewley (1986): Individual uncertainty with fixed interest rate.
I Krusell and Smith (1998): individual and aggregate
uncertainty with GE interest rate.
I Do this using an approximation to the aggregate evolution of
capital.
Heterogeneous Agent Models

I We can write a generic worker’s problem as

1
X
t
max1 E u(ct ) (4)
{ct ,it ,lt }t=0
t=0
s.t. ct + it  rt at + wt lt (5)
at+1 = (1 )at + it (6)
at+1 at (7)
wt ⇠ F (8)
ct 0, lt 0, a0 given (9)

I How we deal with prices rt , wt and choices ct , it , lt is central

to equilibrium.
Recursive Formulation
Wit Stochastic2
.

↓
V (a !, wi
I Can be written as
-V(a
, w)
-
V (a) = u(c) + E [V (a 0 )] (10)
s.t. c + i  ra + wl (11)
a = (1
0
)a + i (12)
a 0
a (13)
w ⇠F (14)
c 0, l 0, a0 given (15)

I Under fairly general conditions, this inherits same properties

as non-stochastic version.
Huggett (1993)

I Endowment economy, no aggregate risk.

I Setup:
I Discrete time;
I Continuum of heterogeneous agents;
I Idiosyncratic endowment risk (labor income stochastic).
I Single bond, a, can be borrowed or saved.
I Borrowling limit, a  0  ait
Idiosyncratic Markov Income Uncertainty

I Suppose wl = e, F [e 0 ] = ⇡(e 0 |e)

I Two states: el , eh
I Can be written as
X
V (a, e) = u(c) + ⇡(e 0 |e)V (a 0 , e 0 ) (16)
e0
s.t. c + a 0  (1 + r )a + e (17)
a 0
a (18)
c 0, a0 given (19)

I Agents want to build precautionary savings again idiosyncratic

risk.
Equilibrium

I Define a distribution of agents over assets as and endowments

e, .
I Stationary equilibrium: aggregate state ( ) is unchanging.
I Agents move around distribution, but LLN ! 0 =
I Define (B) such that given transition function P:
Z
(B) = P(x, B)d (20)
S

I P(x, B) the probability that an agent with state x will have

state B 2 S next period.
I B is a subset of the state space.
Stationary Equilibrium

I Roughly summarizing Huggett, 1993: A stationary equilibrium

for this economy is a tuple (c, a 0 , r , ) that satisfy
1. c and a 0 solve the workers problem taking prices as given.
2. Markets clear:
R R
2.1 consumption = Rproduction: c(x)d = ed
2.2 no net savings: a(x)d = 0
3. is stationary:
Z
(B) = P(x, B)d (21)
S

for all B 2 S
Aiyagari (1994)

I Production economy, no aggregate risk.

I Firms employ capital, households save using capital (really
assets loaned/borrowed from firm).
I Setup:
I Discrete time;
I Continuum of heterogeneous agents;
I Idiosyncratic hours shocks (labor supply stochastic).
I Capital, k, can be borrowed or saved.
I Borrowling limit, k  0  kit
Heterogeneous Agent Production Economy

I In a production economy, the agent’s problem is given by

V (k, ✏; ) = u(c) + E [V (k 0 ✏0 ; 0
)] (22)
s.t. c + k  (1 + r (K , L)
0
)k + w(K , L)✏ (23)
k 0
k (24)
✏ ⇠ MarkovP(✏ |✏) 0
(25)
0
= ( ) (26)
c 0, k 0, k0 given (27)

I ✏ is a markov process that yields hours worked.

I is an unspecified evolution of the aggregate state (k, ✏).
I Prices are determined from the firm’s problem
Prices - The Firm’s Problem

I How we handle prices determines the difficulty of this problem.

I In this economy, a single firm produces using labor (hours)
and capital.

⇧ = max F (K , L) wL rK (28)
K ,L

I This yields standard competitive prices for the rental rates.

Information

I What information do workers need in order to be able to solve

this problem?
I Current period:
I interest rate, r (K , L). This is known from being told the
aggregates at the beginning of the period.
I wage rate, w(K , L). This is known from being told the
aggregates at the beginning of the period.
I Future:
I interest rate and wage rate next period.
I These depend on capital and labor next period.
I Thus, workers need to predict capital and labor in future.
I Rep. Agent model: just need to know their own decision rule.
I Here: need to know distribution across workers, and their
decision rules.
Stationary Recursive Competitive Equilibrium

I A stationary RCE is given by pricing functions r , w, a worker

value function V (k, ✏; ), worker decision rules k 0 , c, a
type-distribution (k, ✏), and aggregates K and L that satisfy
1. k 0 and c are the optimal solutions to the worker’s problem
given prices.
2. Prices are formed competitively from the firm’s problem.
3. Consistency between aggregate evolution and individual
decision rules: is the stationary distribution implied by
worker decision rules.
4. Aggregates
R are consistent
R with individual policy rules:
K = kd , L = ✏d
Return to Capital

I How does return to capital vary by

I serial corr. (⇢) in labor income (think AR1 process)
I and CRRA (µ)?

I Higher ⇢ or µ, more saving, lower return.

Krussell-Smith (1998)

I In the previous model, we relied on the aggregate certainty of

(k, ✏) for a solution by appealing to the law of large numbers.
I i.e., individuals move around the distribution, but those shocks
offset and in the aggregate the distribution is unchanged.
I But what happens if there is aggregate uncertainty?
I Now the distribution changes in the equilibrium, and we need
a way to incorporate this into worker decision rules.
I Krussell-Smith: Aiyagari + aggregate shocks.
I Some details:
I Idiosyncratic labor shock {0,1} markov.
I Aggregate shocks.
I Idiosyncratic shock prob. changes with agg. shocks.
Aggregate Uncertainty
I In a production economy, the agent’s problem is given by

V (k, ✏, z; ) = u(c) + E [V (k 0 ✏0 , z 0 ; 0
)] (29)
s.t. c + k  (1 + r (z, K , L)
0
)k + w(z, K , L)✏ (30)
k 0
k (31)
z = MarkovP(z |z)
0 0
(32)
✏ ⇠ MarkovP(✏ |✏, z ) 0 0
(33)
0
= ( , z, z )
0
(34)
c 0, k 0, k0 given, z0 given (35)

I ✏ is a markov process for employment ✏ 2 {0, 1}

I is an unspecified evolution of the aggregate state.
I z also evolves as a markov process.
I Prices are determined from the firm’s problem.
Prices - The Firm’s Problem

I How we handle prices determines the difficulty of this problem.

I In this economy, a single firm produces using labor (hours)
and capital.

⇧ = max zF (K , L) wL rK (36)
K ,L

I This yields standard competitive prices for the rental rates.

Laws of Motion

I The future aggregate state enters the probability of

employment.
I This means that it impacts all of the laws of motion:

z 0 = MarkovP(z 0 |z) (37)

✏ ⇠ MarkovP(✏ |✏, z ) 0 0
(38)
k  (1 + r (z, K , L)
0
)k + w(z, K , L)✏ c (39)
0
= ( , z, z )
0
(40)

I Because shocks to z change employment status and prices.

Recursive Competitive Equilibrium

I An RCE is given by pricing functions r , w, a worker value

function V (k, ✏, z; ), worker decision rules k 0 , c, a
type-distribution (k, ✏), and aggregates K and L that satisfy
1. k 0 and c are the optimal solutions to the worker’s problem
given prices.
2. Prices are formed competitively from the firm’s problem.
3. Consistency between aggregate evolution and individual
decision rules: is the distribution implied by worker decision
rules given the aggregate state.
4. Aggregates
R are consistent
R with individual policy rules:
K = kd , L = ✏d
Type Distribution

I The type distribution is a problem.

I Each policy function and transition depends on the type
distribution.
I But the type distribution is time-varying in response to
aggregate shocks.
I Alternative: use a smaller number of moments that can be
calculated quickly to characterize the type distribution.
I Like a “sufficient statistic” for the type distribution.
I Discuss the solution to this next time.
Business Cycle Effects

I This model is built to handle stochastic shocks.

I How do heterogeneous agents respond over a business cycle?
Conclusion

I Next time: Solving heterogeneous agent models.

I Schedule: no class on Thursday (4/10)
I Come to conference instead if you can!