EC 744 Lecture Notes: Incomplete Markets and Bewley Models: Jianjun Miao
EC 744 Lecture Notes: Incomplete Markets and Bewley Models: Jianjun Miao
P
a
)=1
j
)
.
Gini Coecient: Two times the area between the Lorenz curve and the
forty ve degree line. It is between zero and one.
Skewness coecient:
P
i
)
i
(a
i
a)
3
cto(a)
3
, where )
i
is the relative
frequency of realization i.
Stylized Facts Read Rodriguez, Diaz-Gimenez, Quadrini and Rios-Rull (2002).
Wealth is the most concentrated, earnings are the second, and the income
is the third.
Wealth is positively correlated with income and earnings, but not strongly.
The movement of households up and down the economic scale is greater
when measured by income than by earnings or wealth
Dierences among these three variables among when the data are disag-
gregated by age, employment status, education, and marital status
The poorest 40% of the population holds a very small amount of wealth
(about 2% of total wealth)
The top 1% holds 30% of total wealth while the top quintile holds almost
80%
Then Gini Index is larger than .78.
2 Markov Chain
Markov chain a
t
o
t=0
with state space S = c
1
, c
2
, ..., c
a
where
1
i)
= Pr
a
t+1
= c
)
[a
t
= c
i
.
Initial probabilities
0
with
0i
= Pr (a
0
= c
i
) .
A set 1 S is called an ergodic set if 1 (c
i
, 1) = 1 for all c
i
1 and
if no proper subset of 1 has this property.
Can compute
Pr
a
t+2
= c
)
[a
t
= c
i
=
a
X
I=1
1
iI
1
I)
Unconditional distributions of a
t
t
1
= Pr (a
1
) =
t
0
1,
t
2
= Pr (a
2
) =
t
0
1
2
,
...
t
I
= Pr (a
I
) =
t
0
1
I
,
where
t
t
= Pr (a
t
)
2.1 Stationary Distribution
Unconditional distributions evolve according to
t
t+1
=
t
t
1
An unconditional distribution is called stationary or invariant if it satises
t
=
t
1
Does
t
converge to ? If yes, and the limit does not depend the initial
distribution
0
, we say the process is asymptotically stationary.
Theorem. Let 1 be a transition matrix with 1
i)
0 for all i, ). Then 1 has
a unique stationary distribution, and the process is asymptotically stationary.
Mean of the stationary distribution
i
1
[a
t
] = j =
a
X
i=1
c
i
i
.
Variance
\ ov
(a
t
) = ov =
a
X
i=1
i
(c
i
j)
2
First-order auto-covariance of the markov chain
Cc (a
t
, a
t1
) =
a
X
i=1
i
a
X
)=1
1
i)
(c
i
j)
c
)
j
.
Exercise Consider a Markov chain with state space c, c and transition
"
j 1 j
1 j j
#
.
Compute its stationary mean, variance and autocorrelation.
3 A Consumption/Savings Problem
Consumption/Savings Problem:
1
2
4
o
X
t=0
o
t
l(c
t
)
3
5
,
subject to
c
t
+o
t+1
= (1 +v)o
t
+&c
t
, o
0
given,
where o (1 +v) < 1.
Borrowing constraint:
o
t+1
o.
DP
(o, c) = max
o
t
o
l((1 +v)o +&c o
t
) +o
Z
(o
t
, c
t
)Q(c, oc
t
)
Policy function o
t
= j (o, c)
Example. Ignore the borrowing constraint for now. Solve for the consump-
tion/savings problem explicitly for (i) CRRA utility without labor income; (ii)
CARA utility with labor income
c
t+1
= jc
t
+.
t+1
where .
t
is IID normal.
3.1 Borrowing Limits: Natural and ad hoc
ad hoc borrowing limit o = b.
natural borrowing limit
o =
&c
1
v
.
Example. Analyze the deterministic consumption/savings model with borrow-
ing constraint for the cases of o (1 +v) < 1, o (1 +v) = 1 and o (1 +v)
1. Please consult the textbook.
3.2 Discrete State Space Method
Discretize the state space
, = [o = o
1
< o
2
... < o
a
], S = [c
1
< c
2
< ... < c
n
]
Discretized Bellman equation
(o
I
, c
i
) = max
o
t
,
l((1 +v)o
I
+&c
i
o
t
) +o
n
X
)=1
1
i)
(o
t
, c
)
).
We obtain discretized value function
Ii
= (o
I
, c
i
)
Step 1. Start with = 0 where is an a n matrix
Step 2. Solve for the above maximization problem to obtain decision rule
and updated value function t
Step 3. If [t [ < ., the stop. Otherwise, let = t and go to step 2.
Suppose the decision rule is given by o
t
= j (o, c)
3.3 Asset-Employment Distributions
Let A
t
(o, c) = Pr (o
t
= o, c
t
= c) . Then
A
t+1
o
t
, c
t
=
X
o
X
c
1
o
t
=j(o,c)
Pr
c
t+1
= c
t
[c
t
= c
A
t
(o, c) .
If there is a solution such that A
t
= A
t+1
= A
+
, then A
+
is called a
stationary distribution.
Interpret A
t
as a distribution of agents over individual state variables (o, c)
Does a staionary distribution exist?
3.4 Long Run Behavior
Euler equation
l
t
(c
t
) o (1 +v) 1
t
h
l
t
(c
t+1
)
i
,
with equality if o
t+1
o.
Let A
t
= o
t
(1 +v)
t
l (c
t
) 0. Then
A
t
1
t
[A
t+1
] .
(A
t
) is a nonnegative supermartingale. By the supermartingale conver-
gence theorem, A
t
converges almost surely to a nonnegative r.v A
+
.
If o (1 +v) 1, then l
t
(c
t
) converges to zero almost surely. So c
t
diverges.
What happens when o (1 +v) < 1? For IID case, Aiyagari (1994) proves
the existence of a stationary distribution. For nite 2 state Markov chain,
Huggett (1993) also proves this result.
Challenging Problem. Can you prove this for general Markov processes?
3.5 Average Assets as function of v
For o (1 +v) < 1, 1o (v) is a continuous function of v!
4 Growth Economies: Aiyagari (1994)
4.1 Goals
Study the impact of heterogeneity and aggregation. This is done by build-
ing a version of the Brock and Mirmon (1972) growth model that allows
for a large number of agents, subject to idiosyncratic risk, who cannot
insure perfectly due to incomplete markets.
Quantify the importance of idiosyncratic risk for savings. Many researchers
have conjectured that precautionary savings may account for a signicant
fraction of aggregate savings.
4.2 The Model
Consumers
A continuum of consumers distributed in [0, 1] with Lebesgue measure
Consumers are ex ante identical, but ex post heterogeneous, having ex-
pected utility
1
2
4
o
X
t=0
o
t
l(c
i
t
)
3
5
, l(c) =
c
1
1
1
,
subject to
c
i
t
+o
i
t+1
= (1 +v
t
)o
i
t
+&
t
c
i
t
, o
i
0
given,
Borrowing constraint:
o
i
t+1
o.
Labor endowment is normalized to 1 and c
i
t
is employment (or earnings)
shock, taking values in a compact set S. For each i, (c
i
t
)
t0
is a Markov
process with a common transition function Q(c, oc).
In steady state, v
t
= v, &
t
= &
c
i
t
+o
i
t+1
= (1 +v)o
i
t
+&c
i
t
, o
i
0
given.
Firm
A single rm with CRTS production function
Y = 1(1, .) = 1
c
1
1c
,
FOC
1
1
(1, 1) = v +c, (1)
1
2
(1, 1) = &. (2)
Stationary Equilibrium
Aggregate distribution
A
t
(1) = (i 1 : (o
i
t
, c
i
t
) 1). (3)
Denition. A stationary (competitive) equilibrium (((o
i
t+1
, c
i
t
)
t0
)
i1
, (v, &), A)
consists of an admissible allocation ((o
i
t+1
, c
i
t
)
t0
)
i1
, a system of prices
(v, &) R
2
+
, and a measure A such that: Given &
t
= & and v
t
= v for
all t 0, then
(i) For -a.e. i, (o
i
t+1
, c
i
t
)
t0
solves the consumers problem.
(ii) The rm maximizes prots so that (1) and (2) are satised, where 1
t
=
R
1
o
i
t
(oi), t 0.
(iii) Markets clear, i.e., for all t 0,
Z
1
c
i
t
(oi) = 1, (4)
C
t
+1
t+1
= 1(1
t
, 1) + (1 c)1
t
, (5)
where C
t
=
R
1
c
i
t
(oi).
(iv) The aggregate distribution is invariant and nonrandom, i.e., A
t
= A a.s.,
where A
t
is given by (3), t 0.
Recursive Equilibrium
Dynamic programming problem
(o, c) = max
o
t
o
l((1 +v)o +&c o
t
) +o
Z
(o
t
, c
t
)Q(c, oc
t
)
A recursive stationary (competitive) equilibrium ((\, j), (v, &), A) consists of
a system of constant prices (v, &) R
2
+
, a value function \ : A S R, a
policy function j : A S A, and a probability distribution A T(A S)
such that:
(i) Given prices (v, &), \ and j solve the above DP.
(ii) Given prices (v, &), the rm maximizes prots, i.e., v = 1
1
(1, 1)c, & =
1
2
(1, 1), where 1 =
R
AS
oA(oo, oc).
(iii) Markets clear:
R
AS
cA(oo, oc) = 1.
(iv) A is an invariant distribution, i.e., for all 1 B(A) B(S),
A(1) =
Z
AS
1
1
c
(v
3
) 1
o
(v
3
)
A
t
j
t1
v
6 Life-Cycle Model: Huggett (1996)
6.1 Goals
Investigate age-wealth distribution at a quantitative level. This is done by
building a version of the Diamond (1965) growth model that allows for (i) a large
number of nitely-lived agents, (ii) earnings, health and longevity uncertainty,
(iii) institutional features such as social security, income taxation, and social
insurance, (iv) market features such as borrowing constraints and the absence
of some insurance markets.
6.2 The Model
Households
OLG Model
Live for . periods, survival probability c
)
conditional on surviving up to
age ) 1.
Population grows at rate a. Age ) agents make up a constant fraction
j
)
= j
)1
c
)
(1 +a) of the population at any time.
Age 1 agent utility
1
2
4
.
X
)=1
o
)
)
i=1
c
i
&
c
)
3
5
where
&(c) = c
1o
(1 o).
Labor endowment in eciency units is given by c(:, )). : is a Markov
shock taking values in a nite set Z. The transition probability is
DP problem
\ (o, :, )) = max
c,o
t
&(c) +oc
)+1
1
h
\ (o
t
, :
t
, ) + 1)[o, :
i
subject to
c +o
t
= o(1 +v(1 t)) + (1 0 t)c(:, ))& +T +b
)
,
c 0, o
t
o and o
t
0 if ) = ..
Social security benet b
)
= 0 for ) 1, = b for ) 1.
Policy function o
t
= j(o, :, ))
Firm
CRTS:
Y = 1(1, 1) = 1
c
1
1c
.
Capital depreciates at rate c
Equilibrium
State space
A = [o, o] Z
Dene
)
(1) as the fraction of age ) agents whose individual states lie in
1 B(A) as a proportion of all age ) agents.
Age distribution
)
(1) =
Z
A
1(o, :, ) 1, 1)
)1
(oo, o:),
where
1(o, :, ) 1, 1) = Pr
:
t
Z : (j(o, :, )), :
t
) 1[:
.
Denition A stationary equilibrium is (c(o, :, )), j(o, :, )), v, &, 1, T, G, b)
and distribution (
1
,
2
, ...,
.
) such that
1. c(o, :, )) and j(o, :, )) are optimal decision rule.
2. Firm optimizes
& = 1
2
(1, 1), & = 1
1
(1, 1).
3. Markets clear
X
)
j
)
Z
A
(c(a, )) +j(a, ))) o
)
+G = 1 (1, 1) + (1 c) 1,
X
)
j
)
Z
A
j(a, ))o
)
= (1 +a)1,
X
)
j
)
Z
A
c(:, ))o
)
= 1 = 1.
4. Distributions are consistent with individual behavior,
)
(1) =
Z
A
1(o, :, ) 1, 1)
)1
(oo, o:),
5. Government budget constraint:
G = t(v1 +&1).
6. Social security benets equal taxes:
0&1 = b
0
@
.
X
)=1
j
)
1
A
.
7. Transfers equal accidental bequests:
T =
2
4
X
)
j
)
(1 c
)+1
)
Z
A
j(a, ))(1 +v(1 t))o
)
3
5
(1 +a).
Calibration
Parameter values
o o c c . 1 c
t
a t 0 o
1.011 1.5 0.8959 0.36 0.06 56,79 46 * 0.012 * 0.1 0,&
Log labor endowment process
j
t
j
t
= (j
t1
j
t1
) +.
t
, .
t
.(0, o
2
.
).
Let :
t
= j
t
j
t
. Approximate j
t
by 18 states nite Markov chain :
t
,
:
t
= :
t1
+.
t
.
Labor endowment
c(:, t) = c
:
t
+j
t
Computation
Step 1. Choose 1.
Step 2. Set & and v according to equilibrium condition 2.
Step 3. Given & and v, nd j(o, :, t) by solving the DP
Step 4. Calculate the wealth distribution and the new capital stock 1
t
.
Step 5. If 1 is approximately equal to 1
t
, stop. Otherwise adjust 1 and
repeat step 2.
6.3 Conclusions
Can replicate measures of both the aggregate wealth and the transfer
wealth in the US economy
Can produce a number of the features of the distribution of wealth in the
US. E.g., match the US wealth Gini and the fraction of wealth held by the
top 20% of US households.
Cannot generate all of the concentration of wealth in the upper tail of the
distribution.
Cannot explain all of the within-age-group inequality.