You are on page 1of 4

Effective risk sharing with 2-sided lack of committment

Kocherlakota 1996 RES


LS Chapter 20
(Note: Approach in class is closer to the paper )
Emperical fact : Conditional on per capita consumption, individual consump-
tion is positively correlated with current and lagged individual income. This is
not Pareto optimal. This may be due to various factors, i.e., moral hazard,
adverse selection etc. Kocherlakota proposes another explation - lack of com-
mittment.
Model: nThere
o are 2 agents (2 types of agents) facing stochastic endowment

streams yi , i = 1, 2.
t=0

For simplicity, assume that yt2 = 1 − yt1 . Denote yt1 = yt ; yt2 = 1 − yt .


n o
yt ∈ y¯1 < . . . < y¯S . Also, assume that y¯1 = 0.

πs = P r{yt = y¯s }, s ∈ {1, 2, . . .}.


Moreover, assume ys and πs are such that the distribution of 1 − yt is the same
as the distribution of ys (in Kocherlakota, yt1 + yt2 = Yt ).
Prob(yt1 = y; yt2 = y 0 ) = Prob(yt1 = y 0 ; yt2 = y). The state of the world
is determined by an iid (over time) random variable θ ∈ {1, 2, . . . , S}, where
Prob(θ = s) = πs .
Denote c1t = ct : consumptn of agent 1; c2t = 1 − ct : consumptn of agent 2
Preferences:
P∞
Eo t=1 β t u(cit ); i =1, 2.
u : u0 > 0, u00 < 0, u ∈ C 2
limc→∞ u0 (c) = +∞ (guarantees interior solution for consumption).
0 < β < 1.
P P
s πs u(y¯s ) πs u(1 − y¯s )
Autarky values: vaut = = s . The second inequality
1−β 1−β
is derived from the symmetric nature of the distribution.
Once uncertainty is realized, agents make simultaneous transfers. We will ana-
lyze the optimal contract. (Self enforcing, Pareto optimal)
Again, using the same arguments as before, we can show that the optimal
contract cannot be Pareto dominated after any history. Therefore, we can write
it as a recursive problem, with the promised value to one of the agents as a state
variable.

1
v : current promised value to agent 1
P(v): corresponding value to agent 2
Bellman Equation:
PS
P(v) = max{cs ,ws }Ss=1 s=1 πs [u(1 − cs ) + βP(ws )]
PS
subject to s=1 πs [u(cs ) + βws ] ≥ v :(Promise Keeping) (µ - LM)
(1) u(cs ) + βws ≥ u(y¯s ) + βvaut .s = 1, . . . S :(PC1 ) (λs - LM)
(2) u(1 − cs ) + βP(ws ) ≥ u(1 − y¯s ) + βvaut , s = 1, . . . S :(PC2 ) (φs - LM)
cs ∈ [0, 1]; ws ∈ [vaut , vmax ], s = 1, . . . S
Langrangian function:
PS PS
L = s=1 πs [u(1 − cs ) + βP(ws )] + µ s=1 πs [u(cs ) + βws − v]
PS
+ s=1 λs [u(cs ) + βws − u(y¯s ) − βvaut ]
PS
+ s=1 µs [u(1 − cs ) + βP(ws ) − u(1 − y¯s ) − βvaut ]
FOC:
(3) cs : −πs u0 (1 − cs ) + µπs u0 (cs ) + λs u0 (cs ) − φs u0 (1 − cs ) = 0
(always have an interior solution as long as u0 (0) = +∞)
(4) ws : πs P0 (ws ) + µπs + λs − φs P0 (ws ) = 0, if ws ∈ (vaut , vmax )
(< 0, if ws = vaut )
(> 0, if ws = vmax )
Claim: It cannot be the case that both λs > 0 and φs > 0 for any s.
Proof: Suppose that λs > 0, φs > 0 for some s. Since, PC1 is binding, we
have: u(cs ) + βws = u(y¯s ) + βvaut and ws ≥ vaut , which implies cs ≤ y¯s .
Similarly, since PC2 is binding we have: u(1 − cs ) + βP(ws ) = u(1 − y¯s ) + βvaut
and P(ws ) ≥ vaut , which implies 1 − cs ≤ 1 − y¯s ⇒ cs ≥ y¯s . But cs ≤ y¯s ≤
cs ⇒ cs = y¯s ⇒ ws = vaut . Also, 1 − cs = 1 − y¯s ⇒P(ws ) = vaut . This is a
contradiction given a non-autarkic sub-game perfect equilibrium exists.
Three cases:
Case 1: λs > 0, φs = 0
Case 2: λs = 0, φs > 0
Case 3: λs = 0, φs = 0
   
0 φs 0 λs
(3) ⇒ u (1 − cs ) 1 + = u (cs ) µ +
πs πs
    
0 φs λs
(4)⇒ P (ws ) 1 + (<, =, >) − µ +
πs πs

2
u0 (1 − cs )
⇒(6) = −P0 (ws ) , if ws ∈ (vaut , vmax ),
u0 (cs )
i.e., neither of vaut ≤ ws ≤ vmax binds.
This gives us a relationship between cs and ws , independent of barys . Because
u(.) and P(.) are concace ⇒ cs = f (ws ), where f (.) is a strictly increasing
function.
Using (5), we get (7):
 
0 φs λs
P (ws ) 1 + = P0 (v) − , if ws ∈ (vaut , vmax )
πs πs
 
φs λs
P0 (ws ) 1 + ≤ P0 (v) − , if ws = vaut
πs πs
 
φs λs
P0 (ws ) 1 + ≥ P0 (v) − , if ws = vmax
πs πs
Case 1: λs > 0; φs = 0
In this case; from (7):
λs
P0 (ws ) = P0 (v) −
πs
P0 (ws ) < P0 (v) ⇒ ws > v as long as v < vmax , if v = vmax , ws = vmax
Case 2: λs = 0; φs > 0
In this case from (7):
 
0 φs
P (ws ) 1 + = P0 (v)
πs
P0 (ws ) > P0 (v) ⇒ ws < v as long as v > vaut , if v = vaut , ws = vaut = v
Case 3: λs = 0 = φs
In this case from (7):
P0 (ws ) = P0 (v) ⇒ ws = v
That is, if agent 1’s PC binds, increase agent 1’s continuation value and con-
sumption (ascs = f (ws )), decrease agent 2’s continuation value and consump-
tion (vice versa if agent 2’s PC binds). If neither PC binds, keep continuation
values and consumption constant for both agents.
As in Thomas and Worrall (1988), if λs > 0 for y¯s , then λ0s > 0 for y¯s0 ≥ y¯s .
Similarly, if φs > 0 for y¯s , then φ0s > 0 for y¯s0 ≤ y¯s .
Important Claim: If in s, s0 , λs > 0, λs > 0 and y¯s0 > y¯s then cs > cs and
ws0 > ws .

3
Proof: Suppose ws0 ≤ ws ⇒ c0s ≤ cs (because cs = f (ws ), f 0 > 0).
u(cs ) + βws = u(y¯s ) + βvaut < u(y¯s0 ) + βvaut = u(cs ) + βws0 ≤ u(cs ) + βws . A
contradiction.
This shows that in states in which PC of an agent is binding, his consumption
and income are positively correlated. If agent 2’s PC are binding, the same is
true for agent 1, the lower the income is, the lower the consumption.
Only in case 3, when neither PC binds, this is not true. That is, consumption
is not correlated with income.
Dynamics: Two cases.
Case 1: First best is a Sub-game Perfect Equilibrium, i.e., part of the constraint
Pareto frontier coincides with the first best Pareto frontier. v∗ , v ∗ are the lowest
and highest value to agent 1 for which this is the case.
Case 2: First best is not a Sub-game Perfect Equilibrium.
Which case actually occurs depend on β. Case 1 occurs for high β and case 2
occurs for low β ⇒ ∃ a threshold β.
Analysis of Case 1: Suppose that at t = 0, start with v0 ∈ [v∗ , v ∗ ] ⇒ full
insurance.
If at t = 0, v0 > v ∗ , then at t = 1, agent 1’s PC doesn’t bind but agent 2’s PC
must bind at some states. To see this, suppose ∃ a s, λs > 0 for t = 1. Then,
ws > v0 > v ∗ ⇒ cs > c∗ (consumption corresponding to v ∗ ).
u(c∗ ) + βv ∗ < u(cs ) + βws = u(ys)
¯ + βvaut
But, this cannot be a part of a SPE because agent 1 has an incentive to deviate.
Now suppose, φs = 0, ∀s at t = 1 ⇒ v0 woulb be the first best, but this
contradicts v0 > v ∗ .⇒ v1 ≤ v0 , (< for some states).
It can be shown that v ∗ ≤ v1 ≤ v0 . By induction, v ∗ ≤ vt+1 ≤ vt ≤ . . . ≤ v0 .
Similarly, if v0 < v ∗ , then v0 ≤ v1 ≤ . . . ≤ v∗ . Implying, there is a monotone
convergence to the first best.
In the long run, convergence to full insurance. The same dynamics hold for
consumption.
Analysis of Case 2: It can be shown that in this case vt and ct convergence to a
stationary distribution, independent of v0 . This distribution is non-degenerate
in the case of s = 2 (see LS). Morover, income and consumption are serially
correlated even in the long run.

You might also like