Lecture Notes in Public Economics∗

Roozbeh Hosseini

August 21, 2009

∗
In preparing these notes I have benefited from: Chari and Kehoe (1998), V.V. Chari and Larry Jone’s
first year Macro lectures notes and V.V. Chari, Narayana Kocherlakota and Mike Golosov’s lectures on
Public Economics. These notes are prepared for the Public Economics course at the W.P. Carey School
of Business, Arizona State University. They contain errors. Please use with caution and do not circulate.
YOU, the reader, are the only person responsible for the errors in the notes! Comments are welcome.

1
ECN 741: Public Economics Fall 2008

Contents

1 Ramsey Taxation - Primal Approach 3

1.1 Ramsey problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Elasticities and optimal taxes . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Additive separable utility functions . . . . . . . . . . . . . . . . . . . 7
1.2.2 Quasi-linear utility function . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Complementarity with leisure . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Uniform commodity taxation . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Intermediate good taxation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Optimal Fiscal Policy-Dynamic Ramsey Taxation 12

2.1 Ramsey problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Chamley-Judd result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Heterogeneous consumers . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Non-Steady State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.3 Werning (QJE, 2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Taxing Capital in Life Cycle Economies (Erosa and Gervais (2002)) . . . . . 26

3 Mirrleesian Approach to Optimal Taxation 30

3.1 The New Dynamic Public Finance . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.1 Inverse Euler Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.2 Long-run properties of efficient allocations . . . . . . . . . . . . . . . 44
3.1.3 Aggregate Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.1.4 Inter-temporal optimality with balanced growth preferences . . . . . 58
3.2 Implementing Efficient Allocations . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.1 Wedges and Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.2 Implemeting Efficient Allocations-General Case . . . . . . . . . . . . 64

References 68

2
ECN 741: Public Economics Fall 2008

1 Ramsey Taxation - Primal Approach

Consider an economy with n types of consumption good that are produced using labor input:

F (c1 + g1 , . . . , cn + gn , l) = 0 (1)

ci is private and gi is public consumption of good i and l is the labor input. F is a constant
return to scale technology. Consumers face the following maximization problem

max U (c1 , . . . , cn , l)
c1 ,...,cn ,l

subject to
n
X
pi (1 + τi )ci = l
i=1

in which τi is the taxed levied on consumption of good i (wage is normalized to 1).

There is a representative firm that produces goods using technology F :

n
X
max p i xi − l
x1 ,...,xn ,l
i=1

F (x1 , . . . , xn , l) = 0

Government has to finance its purchase g = (g1 , . . . , gn ) using linear taxes τi

n
X n
X
pi gi = p i τ i ci (2)
i=1 i=1

Let’s take government purchase as given. A Competitive Equilibrium is

• Consumers and producers allocations: (c, x, l)

• prices: p = (p1 , . . . , pn )

• policy: π = (τ1 , . . . , τn )

such that

1. Given policy π and prices p, (c, l) solve consumers problem.

2. Given prices, p, (x, l) solves producers problem.

3
ECN 741: Public Economics Fall 2008

3. Government budget (equation (2)) holds

4. Allocations are feasible (or market clearing if you like!)

ci + gi = xi for i = 1, . . . , n (3)

Proposition 1 Any competitive equilibrium allocations must satisfy the resource feasibility
constraint
F (c1 + g1 , . . . , cn + gn , l) = 0 (4)

and an implementability constraint

n
X
Ui ci + Ul l = 0. (5)
i=1

Furthermore, any allocations that satisfy (4) and (6) can be supported as a competitive
equilibrium for appropriately constructed polices and prices.

Proof.
Suppose (c, x, l) is a competitive equilibrium allocation. Then the following FOC must hold

Ui
= −(1 + τi )pi for i = 1, . . . , n
Ul

together with the following budget constraint

n
X
pi (1 + τi )ci = l.
i=1

Replacing out for prices (and taxes) from FOC into budget constraint gives the imple-
mentability constraint. The feasibility follows by definition of equilibrium.
Now consider allocations (c, x, l) that are feasible (given vector of g) and satisfy (5). Con-
struct prices from the FOC of the firm

Fi
pi = − for i = 1, . . . , n
Fl

set policy as
Ui Fl
1 + τi = for i = 1, . . . , n
Ul Fi

4
ECN 741: Public Economics Fall 2008

You can verify that the policy and prices (as constructed above) together with the allocation
(c, x, l) is a competitive equilibrium.

We are interested in the problem of choosing the best policy π to maximize the welfare of
consumers. One restriction on such a problem is that the resulting allocation be a competitive
equilibrium allocation for each given policy. The timing is the following: First, government
chooses a policy, Second, private agents makes decision. We are interested in finding the
equilibrium of this game.

1.1 Ramsey problem

Suppose the set of feasible policy for government in Π.

Definition 1 A Ramsey equilibrium is a policy π = (τ1 , · · · τn ) ∈ Π, allocation rules c(·),

x(·) and l(·) and price function p(·) such that

π ∈ arg max U (c(π 0 ), l(π 0 ))

π 0 ∈Π

subject to
n
X n
X
pi gi = p i τ i ci
i=1 i=1

and (c(π ), x(π ), l(π )) together with p(π ) is a competitive equilibrium for every π 0 ∈ Π.
0 0 0 0

Suppose π, (c(·), x(·), l(·)) and p(·) is a Ramsey equilibrium. Then we call (c(π), x(π), l(π))
a Ramsey allocation.

Proposition 2 Suppose c∗ and l∗ are part of a Ramsey allocation. Then

(c∗ , l∗ ) ∈ arg max U (c, l)

c,l

subject to (5) and (4).

Proof.
Follows from the definition of Ramsey allocation.

5
ECN 741: Public Economics Fall 2008

1.2 Elasticities and optimal taxes

Suppose n = 2. Consider the following Ramsey problem

max U (c1 , c2 , l)
c1 ,c2 ,l

subject to

1. Implementability constraint

U1 c1 + U2 c2 + Ul l = 0 (6)

2. Feasibility
F (c1 + g1 , c2 + g2 , l) = 0 (7)

Let λ and γ be multipliers on implementability constraint (equation (6)) and feasibility

(equation (7)). First order conditions are

Ui + λ (Ui + U1i c1 + U2i c2 + Uli l) = γFi i = 1, 2

Ul + λ (Ul + U1l c1 + U2l c2 + Ull l) = γFl

We can write these equations as

Fl
1 + λ − λHl = γ
Ul

in which, Hi = − (U1i c1 +UU2ii c2 +Uli l) and Hl = − (U1l c1 +UU2ll c2 +Ull l) .

Note that from individual problem we have

Ui Fl
1 + τi =
Ul Fi

in other words the optimal wedge must satisfy

1 + λ − λHl
1 + τi =
1 + λ − λHi

There you go! If Hi > Hj , then it is optimal to tax good i more than good j.
The problem is that, it is not very helpful. Unfortunately, without imposing assumption on
U we cannot say much more. Next we consider some special (yet, interesting) cases.

6
ECN 741: Public Economics Fall 2008

1.2.1 Additive separable utility functions

Suppose U is of the form

U (c1 , c2 , l) = u1 (c1 ) + u2 (c2 ) − v(l)

then
Uii ci
Hi = −
Ui
Our goal to relate Hi to income elasticity of demand for good i. In order to do that, suppose
there is a non-wage income m, such that p1 c1 + p2 c2 = l + m. Consider FOC of consumer
(notice that I have ignored taxes for this part)

Ui (ci (p, m)) = pi φ(p, m)

in which φ(p, m) is the lagrange multiplier on budget constrain. Let’s take derivative w.r.t
m
∂ci ∂φ Ui ∂φ
Uii = pi =
∂m ∂m φ ∂m
or
Uii ci m ∂ci m ∂φ
= .
Ui ci ∂m φ ∂m
m ∂ci
Let ηi = ci ∂m
. Then
m ∂φ 1
Hi = −
φ ∂m ηi
Therefore, Hi > Hj if and only if ηj > ηi . Combine this with the above and we get the
following:

Result 1 If preferences are additive separable, necessities should be taxed more than luxuries.

Example : U (c1 , c2 , l) = log(c1 ) + log(c2 − c̄) − v(l)

1.2.2 Quasi-linear utility function

Consider the utility function in the previous section and assume that v(l) = l. Then there
is no income effect and using income elasticities for guiding us about optimal taxes is not
useful. However, we use price elasticities. Consider again the FOC of consumer

Ui (ci ) = pi φ

7
ECN 741: Public Economics Fall 2008

Note that in this case φ = 1 (independent of prices). Take derivative w.r.t pi

∂ci Ui
Uii =φ=
∂pi pi

and
1
Hi =
i

Result 2 If preferences are additive separable and quasi-linear, price-inelastic goods should
be taxed more.

1.2.3 Complementarity with leisure

Sandmo (1987) and Corlett and Hauge (1953-54) argue that goods that are more complement
with leisure should be taxed more heavily. The next example shows this
Example : U (c1 , c2 , l) = cα1 + cα2 (1 − l)β

1.3 Uniform commodity taxation

One of the most useful and interesting result in optimal taxation is the uniform commodity
taxation result. Suppose the preferences are weakly separable in consumption and leisure

U (c1 , . . . , cn , l) = W (G(c1 , . . . , cn ), l) (8)

furthermore, G(·) is homothetic.

Proposition 3 Suppose preferences satisfy (8), then it is optimal to tax all goods at the
same rate, i.e. τi = τj for all i and j.

Proof.
Note that the fact that G(·) is homothetic implies that

Ui (αc, l) Ui (c, l)
=
Uj (αc, l) Uj (c, l)

or
Ui (c, l)
Ui (αc, l) = Uj (αc, l).
Uj (c, l)

8
ECN 741: Public Economics Fall 2008

Differentiate w.r.t αand set α = 1 we get

Pn Pn
k=1 Uik ck k=1 Ujk ck
=
Ui Uj

Also, note that Ul = Wl , Uli = Wlg Gi and Ui = Wg Gi . Therefore,

Pn Pn
k=1 Uik ck Uil l k=1 Uik ck Wlg l
Hi = − − =− − = Hj
Ui Ui Ui Wg

This can be generalized to utility functions of the form

u(c1 , . . . , ck , G(ck+1 , . . . , cn ), l)

in which, G(·) is homothetic. Then the result is that commodities (ck+1 , . . . , cn ) should be
taxed at uniform rate.
Exercise: Suppose consumer is endowed with y unit of good one that cannot be taxed
away. Does the uniform commodity taxation still hold? what if the utility function is
additive separable?
Exercise: Suppose government is restricted to setting taxed on c1 to zero. How would
modify the Ramsey problem? Does the uniform commodity taxation hold?

1.4 Intermediate good taxation

Another powerful and important result in Ramsey taxation is that intermediate good shall
not be taxed.
Suppose there are two sectors. One sector produces commodity x1 that is consumed by
private agent, c1 and by government, g. Commodity x1 is produced using intermediate good
z and labor l1 as input according to the following production function

f (x1 , z, l1 ) = 0.

The other sector, uses labor l2 as input to produce good x2 that can be used as input in
production of good x1 (that is z) or it can be consumed (c2 and g2 ). The technology is the
following

9
ECN 741: Public Economics Fall 2008

h(x2 , l2 ) = 0.

• Private agents solves

max U (c1 , c2 , l1 + l2 )
c,l

subject to
p1 (1 + τ1 )c1 + p2 (1 + τ2 )c2 ≤ l1 + l2 .

• Producer of good x1 solves

max p1 x − l1 − p2 (1 + τz )z
x1 ,z,l1

subject to
f (x1 , z, l1 ) = 0.

The FOC for this problem implies

fz
= p2 (1 + τz ).
fl

• Producer of good x2 solves

max p2 x2 − l2
x2 ,l2

subject to
h(x2 , l2 ) = 0.

and FOC implies

hx
= −p2 .
hl
Combining the FOC condition for two sector we get

hx fz
(1 + τz ) = − .
hl fl

• Government budget constraint is

τ1 p1 c1 + τ2 p2 c2 + τz p2 z = p1 g1 + p2 g2

10
ECN 741: Public Economics Fall 2008

• Finally, feasibility and market clearing

c1 + g1 = x1
c2 + g 2 + z = x 2
f (x1 , z, l1 ) = 0
h(x2 , l2 ) = 0

The Ramsey problem is

max U (c1 , c2 , l1 + l2 )

subject to

U1 c1 + U2 c2 + Ul (l1 + l2 ) = 0 λ
f (c1 + g1 , z, l1 ) = 0 φ1
h(c2 + g2 + z, l2 ) = 0 φ2

FOC w.r.t z
φ1 fz = −φ2 hx

FOC w.r.t l1 and l2

Ul + λ(Ull (l1 + l2 ) + Ul + Ucl c) = fl φ1

Ul + λ(Ull (l1 + l2 ) + Ul + Ucl c) = hl φ2

and therefore,
fl φ1 = hl φ2 .

This implies that

hx fz
=−
hl fl
It means that it is optimal to set τz = 0 and not distort production efficiency. For more on
intermediate good taxation and production efficiency see Diamond and Mirrlees (1971).

11
ECN 741: Public Economics Fall 2008

2 Optimal Fiscal Policy-Dynamic Ramsey Taxation

The main focus of this section is the derivation of Chamley-Judd result (Chamley (1986)
and Judd (1985)). We are only going to consider deterministic environment. See Chari and
Kehoe (1994) and Chari and Kehoe (1998) for stochastic environment and optimal policy
over business cycle.
The environment is the following. There are infinitely lived identical consumers. Government
has to finance expenditure gt every period and levies distortionary taxes (or subsidies) on
consumption, investment, labor and capital income. It can also issue debt.
Consumer’s problem: consumers are endowed with k0 unit of capital and b0 unit of govern-
ment debt

∞
X
max β t U (ct , lt )
ct ,lt ,xt ,kt+1 ,bt+1
t=0

subject to

(1 + τct )ct + (1 + τxt )xt + bt+1 ≤ (1 − τlt )wt lt + (1 − τkt )rt kt + Rbt bt ; λt
kt+1 ≤ (1 − δ)kt + xt
−bt+1 ≤ M

k0 , b0 given

in which M is some large positive number.

The FOC’s are

β t Uct = λt (1 + τct ) (9)

−β t Ult = λt wt (1 − τlt ) (10)
(1 + τxt )λt = λt+1 [(1 − τxt+1 )(1 − δ) + (1 − τkt+1 )rt+1 ] (11)
λt = λt+1 Rbt+1 (12)

Government Budget:

gt + Rbt bt = bt+1 + τxt xt + τct ct + τlt wt lt + τkt rt kt (13)

Feasibility:
ct + gt + kt+1 = F (kt , lt ) + (1 − δ)kt (14)

12
ECN 741: Public Economics Fall 2008

Competitive pricing implies that

rt = Fk (kt , lt ) (15)
wt = Fl (kt , lt )

A competitive equilibrium is: the sequence of allocations x = {ct , lt , bt+1 , kt+1 , xt }∞ t=0 , prices
∞ ∞
{rt , wt , Rbt }t=0 , policy π = {τct , τlt , τxt , τkt+1 }t=0 such that, the allocations solve consumer
problem, given prices and policy, prices are competitive, government budget holds and allo-
cations are feasible.
A Ramsey Equilibrium is a policy π, an allocation rule x(·) and price rules r(·), w(·) and
Rb (·) such that:
X∞
π ∈ arg max β t U (ct , lt )
t=0

subject to 12 and x(π) be a competitive equilibrium, and

for any policy π 0 , allocation x(π 0 ) and prices (r(π 0 ), w(π 0 ), Rb (π 0 )) be a competitive equilib-
rium.
We next derive the implementability condition. Note that if conditions of Ekeland and
Scheinkman (1986) and/or Weitzman (1973) are satisfied, then the equilibrium allocations
should also satisfy the following Transversality conditions

lim λt bt+1 = 0 (16)

t→∞
lim λt kt+1 = 0 (17)
t→∞

Now multiply consumer’s budget constraint by λt and sum over t and use (16)-(17)
∞
X ∞
X
λt [(1 + τct )ct + (1 + τxt )(kt+1 − (1 − δ)kt ) + bt+1 ] = λt [(1 − τlt )wt lt + (1 − τkt )rt kt + Rbt bt ] .
t=0 t=0

Now use (9)-(12) and we get

∞
X
λt [(1 + τct )ct − (1 − τlt )wt lt ] = λ0 {[(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k0 + Rb0 b0 } .
t=0

13
ECN 741: Public Economics Fall 2008

Now replace (9)-(10) and we arrive at the implementability constraint

∞
X
β t [Uct ct + Ult lt ] = U0 {[(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k0 + Rb0 b0 } (18)
t=0

Proposition 4 A feasible allocation x = {ct , lt , bt+1 , kt+1 , xt }∞

t=0 is a competitive equilibrium
allocation if and only it satisfies the implementability constraint (18) (for some period zero
policies).

Proof.
Suppose x is the competitive equilibrium allocation, then following the steps outlines above
we can show that it should satisfy the implementability constraint (18). Now suppose an
allocation x∗ is feasible and satisfy (18) for some proof zero policies.
Note that in any competitive equilibrium, the bond holding must satisfy
∞
X [Ucs cs + Uls ls ]
bt+1 = β t−s − kt+1 (19)
s=t+1
Uct

∗
in other words, any sequence of c∗t , lt∗ and kt+1 uniquely identifies a sequence of bt that is a
part of competitive equilibrium. Candidate wage and rate of return on capital is given by
(15). Therefore, from the FOC (9)-(12) we have

1 − τlt U∗
= − ∗ lt ∗
1 + τct Flt Uct
∗ ∗
U Uct+1 ∗
(1 + τxt ) ct
 
= β (1 − τxt+1 )(1 − δ) + (1 − τkt+1 )Fkt+1 (20)
1 + τct 1 + τct+1
∗ ∗
Uct Uct+1
= β Rbt+1
1 + τct 1 + τct+1

any two of the four taxes can be chosen such that the above conditions hold.

2.1 Ramsey problem

The Ramsey problem is the following

∞
X
max β t U (ct , lt )
ct ,kt+1 ,lt
t=0

14
ECN 741: Public Economics Fall 2008

subject to
∞
X
β t [Uct ct + Ult lt ] = U0 {[(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k0 + Rb0 b0 } ;λ
t=0

ct + gt + kt+1 = F (kt , lt ) + (1 − δ)kt ; φt

Define function W (·, ·, ·) as

W (c, l, λ) = U (c, l) + λ [Uc c + Ul l] .

Now we can rewrite the Ramsey problem as

∞
X
max β t W (ct , lt , λ)
ct ,kt+1 ,lt
t=0

subject to
ct + gt + kt+1 = F (kt , lt ) + (1 − δ)kt ; φt

Take first order conditions

Wlt
= −Flt (21)
Wct
Wct
= β(1 − δ + Fkt ) for t ≥ 1 (22)
Wct+1

2.2 Chamley-Judd result

Proposition 5 If the solution to the Ramsey problem converges to a steady state, then at
the steady state, the tax rate on capital income is zero.

Proof.
In (22) at the steady state we have

β(1 − δ + Fkt+1 ) = 1.

This implies that at the steady state there is no inter-temporal distortion. Compare with
(20) we have
   
(1 + τxt )(1 + τct+1 ) 1 − τkt+1
=β 1−δ+ Fkt+1
(1 + τct )(1 + τxt+1 ) 1 + τxt+1

15
ECN 741: Public Economics Fall 2008

Note that any feasible allocation that satisfies (18) can be implemented by two of the four
taxes (that is we only need two of the τc , τl ,τx and τk to implement the same allocations).
This in turn implies that

τkt = 0
1 + τct
= constant
1 + τxt

2.2.1 Heterogeneous consumers

Suppose there are two type of consumers i = 1, 2 with preferences

∞
X
β t U i (cit , lit )
t=0

The resources constraint for the economy is

c1t + c2t + kt+1 = F (kt , l1t , l2t ) + (1 − δ)kt (23)

implementability constraint for consumer i is

∞
X
β t Ucti cit + Ulti lit = U0i [(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k0i + Rb0 bi0
  
(24)
t=0

Suppose government puts welfare weights ωi on consumers of type i. The Ramsey problem
is

∞
X ∞
X
t 1
max ω1 β U (c1t , l1t ) + ω2 β t U 2 (c2t , l2t )
t=0 t=0

subject to (23) and (24).

Attached multiplier λi to implementability constraint of type i and write
X
ωi U i (ci , li ) + λi Uci ci + Uli li


W (c1 , c2 , l1 , l2 , λ1 , λ2 ) =
i=1,2

16
ECN 741: Public Economics Fall 2008

∞
X
max β t W (c1t , c2t , l1t , l2t , λ1 , λ2 )
t=0

subject to
c1t + c2t + kt+1 = F (kt , l1t , l2t ) + (1 − δ)kt ; φt

where W i is defined the obvious way.

First order conditions imply

Wcit = βWcit+1 (1 − δ + Fkt+1 )

and in the steady state

1 = β(1 − δ + Fkt+1 )

and, therefore, tax on capital should be zero in the steady state.

Capitalists vs Workers (Judd 1985)

Suppose consumer of type 1 does not hold any asset and cannot save, borrow or invest. We
call these ’Worker’. Also, assume that all the capital is held by consumer 2 who do not
supply any labor. We call these ’Capitalists’. The implementability constraint for ’Worker’
is
Uct1 c1t + Ult1 l1t = 0 ∀ t

and for ’Capitalist’

∞
X
β t Uct2 c2t = U02 [(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k02 + Rb0 b20
  
(25)
t=0

Suppose the welfare weight on ’Worker’ utility is 1 and on ’Capitalist’ utility is zero.

∞
X
max β t U 1 (c1t , l1t )
t=0

subject to
Uct1 c1t + Ult1 l1t = 0 ∀ t
∞
X
β t Uct2 c2t = U02 [(1 + τx0 )(1 − δ) + (1 − τk0 )r0 ] k02 + Rb0 b20

(26)
t=0

c1t + c2t + kt+1 = F (kt , l1t , l2t ) + (1 − δ)kt ; φt

17
ECN 741: Public Economics Fall 2008

Define
W (c1 , c2 , l1 , l2 , λ1 , λ2 ) = U 1 (c1 , l1 ) + λi Uci ci + Uli li

First order conditions

λβ t Ucct
 2
c2t + Uct2 + φt = 0


φt = φt+1 (1 − δ + Fkt+1 )

in steady state φt+1 = βφt and therefore

1 = β(1 − δ + Fkt+1 )

and again, tax of capital is zero in the steady state.

Exercise: In the above set up we have implicitly assumed that government can levy different
taxes on different consumer types. How would you add the following restrictions to the
problem

1. Tax on capital income has to be uniform across different types. Does the result hold
with this restriction? Under what assumptions?

2. Tax on labor income has to be uniform across different types. Does the result hold?
Under what assumptions?

3. Tax on capital income cannot be more than 100 percent. Does the result hold? Under
what assumptions?

Dividend Taxes?!!! (an interesting example)

Suppose we write the environment as in McGrattan and Prescott (2005) with corporate taxes
and dividend taxes. Consumers can trade share of corporations, st , at price vt . Let dt be
dividend and τdt be dividend tax. Consumers solve
∞
X
max β t U (ct , lt )
ct ,st+1 ,lt
t=0

subject to
∞
X ∞
X
pt [ct + vt (st+1 − st )] ≤ pt [(1 − τdt )dt st + (1 − τlt )wt lt ]
t=0 t=0

s0 = 1

18
ECN 741: Public Economics Fall 2008

FOC implies
Uct
= −(1 − τlt )wt
Ult
pt vt = pt+1 vt+1 + pt+1 (1 − τdt+1 )dt+1

And therefore implementability constraint is

∞
X
β t [Uct ct + Ult lt ] = Uc0 [v0 + (1 − τd0 )d0 ] s0 (27)
t=0

There is a corporation that maximizes the present discounted value of owners’ dividends and
pays taxes τt on corporate income.

∞
X
max pt (1 − τdt )dt
t=0

subject to
dt = f (kt , lt ) − xt − wt lt − τt (f (kt , lt ) − δkt − wt lt )

kt+1 = (1 − δ)kt + xt

First order conditions for the corporation is

flt = wt

pt (1 − τdt )
= 1 − (1 − τt+1 )(fkt+1 − δ)
pt+1 (1 − τdt+1 )

For this economy the feasibility is

ct + kt+1 + gt = f (kt , lt ) + (1 − δ)kt

st = 1

and there is also a government budget constraint

∞
X ∞
X
pt gt = pt [τdt dt st + τt (f (kt , lt ) − δkt − wt lt ) + τlt wt lt ]
t=0 t=0

Question: What is the appropriate implementability constraint? Is constraint (27) suffi-

cient? In other words, is it true that any feasible allocation that satisfy (27) can be supported
in a competitive equilibrium? If not, what other constraints should be added?

19
ECN 741: Public Economics Fall 2008

2.2.2 Non-Steady State

Proposition 6 Suppose the utility function is of the form

c1−σ
U (c, l) = − v(l),
1−σ
Then Ramsey taxes on capital income is zero for t ≥ 2.

Proof.
Do it as an exercise.

Exercise: Can you establish any connection between this result and uniform commodity
taxation?

2.2.3 Werning (QJE, 2007)

Werning (2007) studies a dynamic environment in which individuals are heterogeneous in

their skills. Instead of ruling out lump-sum taxation, he allows them. However, he does not
allow government to condition the lump-sum tax on individual skill. Instead he allows for a
distortionary labor income (and capital income) tax that government can use to redistribute
income across people with different skill. In some sense, it is one step away from traditional
Ramsey setup, towards rationalizing distortionary taxes.
The environment is the following: let c be consumption and l be the hours worked. Individual
with skill θ who works l hours produce y = θl efficiency labor unit. If period utility over
hours worked and consumption is U (c, l), then we can write it in terms of consumption and
efficiency labor unit as U i (c, y) = U (c, y/θi ).

Suppose there are θ ∈ Θ = θi , . . . , θN . We call the individual of type θi , person i or type
i. The fraction of type i is π i . Assume i π i θi = 1.
P

Aggregate state of economy is st ∈ S (finite set) and is publicly observable. Denote the
history of aggregate shocks by st = (s0 , . . . , st ). Probability of history st is Pr(st ).
Consumer problem
Individual of type i solves
X
max β t Pr(st )U i (ct (st ), y(st )) (28)
t,st

20
ECN 741: Public Economics Fall 2008

sub. to
X  X
p(st ) c(st ) + k(st ) ≤ p(st ) wt (st )(1 − τ (st ))y(st ) + R(st )k(st−1 ) − T
  

t,st t,st

k i (s0 ) = k0i is given

in which R(st ) = 1 + (1 − κ(st ))(rt (st ) − δ) and T = t,st p(st )T (st ) is present value of
P

lump-sum taxes. Note that there is heterogeneity in skills θi as well as initial capital holding
k i (s0 ).
Feasibility
Let L(st ) = π i y i (st ), C(st ) = π i ci (st ), K(st ) = π i k i (st ). Then feasibility is
P P P
i i i

C(st ) + K(st ) + g(st ) = F (K(st−1 ), L(st ), st , t) + (1 − δ)K(st−1 ) (29)

Govern met
Government has exogenously given sequence of expenditure g(st ) to finance. It can levy
linear tax on capital income κ(st ). It can also levy the following tax on income

τ (st )wt (st )y i (st ) + T (st )

Government budget constraint is

X X
p(st )g(st ) ≤ T + p(st ) τ (st )wt (st )L(st ) + κ(st )(rt (st ) − δ)K(st−1 )
 
(30)
t,st t,st

F irms
As usual the firm’s problem is static and implies marginal product pricing

rt (st ) = Fk (K(st−1 ), L(st ), st , t) (31)

wt (st ) = FL (K(st−1 ), L(st ), st , t)

Equilibrium is defined the usual way.

Next we derive the implementability constraints. Werning (2007) develops a methodology
that incorporates the fact that labor income taxes are uniform across types (so no extra con-
straint needs to be added to the optimal taxation problem). Also, he shows implementability
constraints can be written only in terms of aggregates.

21
ECN 741: Public Economics Fall 2008

First observe that in any equilibrium

Uyi (st ) Uyj (st ) t t

= j t = −w(s )(1 − τ (s )) (32)
Uci (st ) Uc (s )
i t
Uc (s ) Ucj (st ) p(st )
= = ∀ i, j
Uci (s0 ) Ucj (s0 ) β t Pr(st )p(s0 )

Therefore, given the aggregate consumption and labor output (C(st ), L(st )), the assignment
of allocation of consumption and labor output {ci (st ), y i (st )} are efficient. In other words,

given any sequence of aggregate output (C(st ), L(st )), there are weights ϕ = ϕ1 , . . . , ϕN
such that i π i ϕi = 1 and {ci (st ), y i (st )} is the solution to
P

X
U m (C(st ), L(st ); ϕ) ≡ max
i i
π i ϕi U i (ci , y i )
{c ,y }

sub. to
X X
π i ci = C(st ), π i y i = L(st )
i i

Denote the solution by

ci = hic (C, L; ϕ), y i = hiy (C, L; ϕ) (33)

therefore
(ci (st ), y i (st )) = hi (C, L; ϕ)

in which hi = (hic , hiy ).

Note also that

UCm (C(st ), L(st ); ϕ) = ϕi Uci (ci , y i ) (34)

ULm (C(st ), L(st ); ϕ) = ϕi Uyi (ci , y i )

and therefore, in any equilibrium

ULm (st )
= −w(st )(1 − τ (st )) (35)
UCm (st )
Ucm (st ) p(st )
= ∀ i, j
Ucm (s0 ) β t Pr(st )p(s0 )

Now let’s look at individual i’s implementability constraint

X
β t Uci (ci (st ), y i (st ))ci (st ) + Uyi (ci (st ), y i (st ))y i (st ) = Uci (ci (s0 ), y i (s0 )) R0 k0i − T
   

t,st

22
ECN 741: Public Economics Fall 2008

Now we can replace individual i’s allocations in terms of aggregate allocations using (33)
and (34)

X
β t UCm (C(st ), L(st ); ϕ)hic (C(st ), L(st ); ϕ) =

(36)
t,st

+ULm (C(st ), L(st ); ϕ)hiy (C(st ), L(st ); ϕ) = Uci (C(s0 ), L(s0 ); ϕ) R0 k0i − T ∀i(37)
  

Note that (36) is expressed entirely in terms of aggregate allocations, weights ϕ and initial
endowments.

Proposition 7 Given initial wealth R0 k0i , an aggregate allocation {C(st ), L(st ), K(st )} can
be implemented in a competitive equilibrium if and only if

1. It is feasible

2. There exists weights ϕ and lump-sum T such that implementability constraint (36)
holds for all i = 1, . . . , N

Proof.
Any equilibrium allocation is feasible and we just showed that it satisfy (36) . Suppose there
is a feasible aggregate allocation that satisfies (36) for sum weights and lump-sum taxes.
Then individual allocations and prices can be constructed using (33) and (35). Then it is
immediate that (32) (consumer optimality) holds. The individual allocations constructed as
such are also feasible since they satisfy (36) .

A Panning Problem

Suppose λi is planer’s weight on type i.

P i i
i π λ = 1. Consider the following planning
problem
X
max λi π i β t Pr(st )U i (hi (C(st ), L(st ); ϕ))
t,st ,i

sub. to
X
β t UCm (C(st ), L(st ); ϕ)hic (C(st ), L(st ); ϕ)


t,st

+ULm (C(st ), L(st ); ϕ)hiy (C(st ), L(st ); ϕ) = Uci (C(s0 ), L(s0 ); ϕ) R0 k0i − T ∀i ; µi π i
  

23
ECN 741: Public Economics Fall 2008

C(st ) + K(st ) + g(st ) = F (K(st−1 ), L(st ), st , t) + (1 − δ)K(st−1 )

Make our usual change of variable

X
W (C, L; ϕ, µ, λ) ≡ π i λi U i (hi (C, L; ϕ))
i
+µi UCm (C, L; ϕ)hic (C, L; ϕ) + ULm (C, L; ϕ)hiy (C, L; ϕ)
 

and rewrite the problem as

X X
λi π i β t Pr(st )W (C(st ), L(st ); ϕ, µ, λ) − Uci (C(s0 ), L(s0 ); ϕ) π i µi R0 k0i − T
 
max
t,st ,i i

sub. to

C(st ) + K(st ) + g(st ) = F (K(st−1 ), L(st ), st , t) + (1 − δ)K(st−1 )

First order conditions are

WC (C(st ), L(st ); ϕ, µ, λ)
FL (K(st−1 ), L(st ), st , t) = −
WL (C(st ), L(st ); ϕ, µ, λ)
X
WC (C(st ), L(st ); ϕ, µ, λ) = β WC (C(st+1 ), L(st+1 ); ϕ, µ, λ)R∗ (st+1 )Pr(st+1 )
st+1 |st

in which R∗ (st+1 ) = 1 + δ + FK (K(st ), L(st+1 ), st+1 , t + 1).

FOC with respect to tax on initial capital
X
µi π i k0i = 0 or R0 = 0
i

Optimal Taxes

ULm (C, L; ϕ) WC (C, L; ϕ, µ, λ)

τ ∗ (st ) = 1 −
WL (C, L; ϕ, µ, λ) UCm (C, L; ϕ)

Consumer inter-temporal optimality in equilibrium implies

X
UCm (C(st ), L(st ); ϕ) = β UCm (C(st+1 ), L(st+1 ); ϕ)R(st+1 )Pr(st+1 )
st+1 |st

24
ECN 741: Public Economics Fall 2008

One way to get this is to set the capital income taxes such that

UCm (C(st ), L(st ); ϕ) WC (C(st+1 ), L(st+1 ); ϕ, µ, λ)

R(st+1 ) = R∗ (st+1 )
WC (C(st ), L(st ); ϕ, µ, λ) UCm (C(st+1 ), L(st+1 ); ϕ)

Note that FOC with respect to initial capital implies

N
X
µi k0i π i = 0
i=1

Example: Consider the following preferences

γ
i c1−σ (y/θi )
U (c, y) = −α
1−σ γ

Note that we have hic (C, L; ϕ) = ωci C and hiy (C, L; ϕ) = ωyi L, with
γ −1
1/σ
(ϕi ) (θi ) γ−1 (ϕi ) γ−1
ωci =P and ωyi =P γ −1
i 1/σ
i πi (ϕ ) i πi (θi ) γ−1 (ϕi ) γ−1

and therefore
γ γ
m c1−σ i
m (y/θ ) W c
1−σ i
W (y/θ )
U = Φm
u − Φv α and W = Φu − Φv α
1−σ γ 1−σ γ

in which Φm m W W
u , Φv , Φu and Φv are some constant. Note that this implies

Φm W
v Φu
τ ∗ (C, L) = 1 −
Φm W
u Φv

Note also that

UCm (C(st ), L(st ); ϕ) WC (C(st+1 ), L(st+1 ); ϕ, µ, λ)

=1
WC (C(st ), L(st ); ϕ, µ, λ) UCm (C(st+1 ), L(st+1 ); ϕ)

and therefore
R(st+1 ) = R∗ (st+1 )

which implies
κ(st ) = 0 for all t ≥ 1

This implies that the result for optimal taxes on capital income holds from date zero (not
just for t ≥ 1 as it was the case before). When k0i = k0 for all i, taxing initial capital is

25
ECN 741: Public Economics Fall 2008

like a lump-sum tax. But since lump-tax is allowed here, it is not necessary. However, when
individuals are heterogeneous in their initial wealth, then taxing wealth for redistribution is
desirable.
Example: Now consider the following preferences

y
U i (c, y) = α log(c) + (1 − α) log(1 − )
θi

then hic (C, L; ϕ) = ω i C and hiy (C, L; ϕ) = θi − ω i (1 − L) and

ϕi
ωi = P i i
iπ ϕ

therefore,
X
U m (C, L; φ) = α log(C) + (1 − α) log(1 − L) + α log ω i + (1 − α) log ω i /θi .



i

Also we can we can verify that

(1 − α)
W (C, L) = ΦW W
U (α log(C) + (1 − α) log(1 − L)) + ΦUL
1−L

and therefore
1
τ ∗ (L) =
(1 − L)ΦW W
U /ΦUL +1
also
κ(st ) = 0 for all t ≥ 1

2.3 Taxing Capital in Life Cycle Economies (Erosa and Gervais

(2002))

Here, I present a 2 period version of Erosa and Gervais (2002). Individuals live 2 periods
(born at age 0, die at age 1). Each generation is indexed by its date of birth. For example
in period t, the generations alive are t − 1, t. Assume no population growth.
Each individual is endowed with one unit of time at each age j and can transform one unit
of time into zj unit of efficient labor. Let ct,j be the consumption of generation t at age j.
Other variables follow the same notation.

26
ECN 741: Public Economics Fall 2008

Consumer’s problem is the following (for generation t > 0)

max U (ct,0 , lt,0 ) + βU (ct,1 , lt,1 )

subject to

c l
(1 − τt,0 )ct,0 + at,1 ≤ (1 − τt,0 )wt z0 lt,0
c l k
(1 − τt,1 )ct,1 ≤ (1 − τt,1 )wt z1 lt,1 + (1 + (1 − τt,1 )(rt − δ))at,1

There is a constant return to scale technology and

rt = fk (kt , lt )
wt = fl (kt , lt )

and feasibility requires that

ct + kt+1 = f (kt , lt ) + (1 − δ)kt

ct = ct,0 + ct−1,1
lt = lt,0 + lt−1,1
kt = at−1,1

Government budget constrain is

∞ ∞
" #
X X X X
c l k
pt gt = pt τt−j,j ct−j,j + τt−j,j wt zj lt−j,j + τt−1,1 (rt − δ)at−1,1
t=0 t=0 j=0,1 j=0,1

Let U t = U (ct,0 , lt,0 ) + βU (ct,1 , lt,1 ) be the lifetime utility of generation t for a given se-
quence of consumption and leisure and let 0 < γ < 1 be government’s discount factor across
generations. Government objective is to maximize
∞
X
γ tU t
t=0

Exercise: Show that, in this environment, implementability constraint for generation t is

the following

Uct,0 ct,0 + Ult,0 lt,0 + β Uct,1 ct,1 + Ult,1 lt,1 = 0 (38)

27
ECN 741: Public Economics Fall 2008

Exercise: Show that a feasible allocation is implementable if and only if it satisfy (38).

Ramsey problem

Ramsey problem is the following

∞
X
max γ t [U (ct,0 , lt,0 ) + βU (ct,1 , lt,1 )]
t=0

subject to
Uct,0 ct,0 + Ult,0 lt,0 + β Uct,1 ct,1 + Ult,1 lt,1 = 0 ; γ t λt

ct + kt+1 = f (kt , lt ) + (1 − δ)kt ; γ t φt

ct = ct,0 + ct−1,1
lt = lt,0 + lt−1,1
kt = at−1,1

First order conditions are

γ t Uct,0 + γ t λt Uct,0 + Ucct,0 ct,0 + Ulct,0 lt,0 = γ t φt

(39)
γ t βUct,1 + γ t βλt Uct,1 + Ucct,1 ct,1 + Ulct,1 lt,1 = γ t+1 φt+1

γ t Ult,0 + γ t λt Ult,0 + Ullt,0 lt,0 + Ulct,0 ct,0 = γ t φt flt

(40)
γ t βUlt,0 + γ t βλt Ult,1 + Ullt,1 lt,1 + Uclt,1 ct,1 = γ t+1 φt+1 flt+1

γ t φt = γ t+1 φt+1 (1 − δ + fkt+1 ) (41)

Combine (39) and (41)

Uct,0 + λt Uct,0 + Ucct,0 ct,0 + Ulct,0 lt,0

= β (1 − δ + fkt+1 ) (42)
Uct,1 + λt Uct,1 + Ucct,1 ct,1 + Ulct,1 lt,1

Steady State: In the steady state (ct,0 , ct,1 , lt,0 , lt,1 , at,1 ) = (c0 , c1 , l0 , l1 , a1 ) and λt = λ.
Therefore,

28
ECN 741: Public Economics Fall 2008

Uc0 + λ (Uc0 + Ucc0 c0 + Ulc0 l0 )

= β (1 − δ + fkt+1 )
βUc1 + λ (Uc1 + Ucc1 c1 + Ulc1 l1 )

Note that this in general does not imply zero tax on capital. When profile of labor produc-
tivity, zj , is not flat over lifetime, in general consumption and leisure allocations over lifetime
is not flat.
Question: Intuitively, why is it optimal to distort inter-temporal decision in this environ-
ment?
We can impose assumptions on preferences (both for government and individuals) to arrive
at zero capital taxation result again.

Proposition 8 Suppose period utility function is of the following form

c1−σ
u(c, l) = − v(l)
1−σ

Then the Ramsey problem prescribes no inter-temporal distortions for periods t ≥ 1, provided
that labor income taxes can be age-dependent.

Proof.
Note that equation (42) becomes

Uc0,t
= β (1 − δ + fkt+1 )
Uc1,t

Individual problems Euler equation is

Uc0,t
= β (1 + (1 − τt,1 )(fkt+1 − δ))
Uc1,t

This result should be viewed as a consequence of uniform commodity taxation.

Question: Note that we get this result independent of γ. Isn’t that surprising? Why is
that?

29
ECN 741: Public Economics Fall 2008

3 Mirrleesian Approach to Optimal Taxation

One criticism of the Ramsey approach is the ad hoc assumption of linear distortionary taxes
exclusion of Lump-sum taxation. At the same time without imposing these restrictions or
without including informational frictions the Lump-taxes are very desirable in these models.
We saw Werning (2007) as an attempt to move away from this limitations and expand
the set of instruments available to government, i.e., Lump-sum taxes together with uniform
distortionary taxes across individuals. In doing that he appeals to informational friction, i.e.,
the fact that individuals type (skill) is not observable and hence cannot be taxes. But is it
the best government can do? Is it possible for government to implement more sophisticated
instruments and achieve “better” outcomes (lets agree for now that “better” means, higher
welfare given a welfare function)? What are these instruments? How we possibly restrict
the set instruments available to government?
In Mirrleesian approach set of instruments is pinned down by information/enforcement lim-
itation of the government. The optimal tax policy is found among those policies that are
incentive compatible, given the information/enforcement restrictions. In doing that we pro-
ceed in two steps

1. Find a socially optimal allocation given information/enforcement restrictions

2. Devise a tax system that implements this allocation

Environment with finite number of agents

There are N agents, indexed by n = 1, . . . , N , who live T < ∞ period. The preferences are

T
X
β t−1 (u(ct ) − v(lt )) , 1 > β > 0, u0 , −u00 , v 0 , v 00 > 0
t=1

in which ct is consumption and is observed. lt is hours worked (or effort) and it is not
observed.
Let Θ be a finite set of skills or ability. Nature makes a draw θnT = (θn1 , . . . , θnt ) ∈ ΘT =
Θ × · · · × Θ for each agent n. The θnT draws are i.i.d across agents. Let π(·) be probability
density function over ΘT draws. Each agent privately learns θnt at the beginning of period
t. Individual who has skill θnt and works lnt hours in period t can produce ynt unit of output
according to
ynt = lnt · θnt

30
ECN 741: Public Economics Fall 2008

lnt and θnt are both private information but ynt is observable. In what follows we make a
change of variable lnt = yθnt
nt
.

Definition 2 An allocation is a sequence of functions (cn , yn )N

n=1

N
cn : ΘT −→ RT+

N
yn : ΘT −→ RT+

such that cnt and ynt are (θ1t , . . . , θN

t
) − measurable. Denote the set of feasible allocation by
F A.

Definition 3 An allocation (cn , yn )N

n=1 is feasible if

T X
X N T X
X N
cnt (θ1T , . . . , θnT )R−t ≤ ynt (θ1T , . . . , θnT )R−t , R>1
t=1 n=1 t=1 n=1

for all (θ1T , . . . , θnT ) such that π(θ1T , . . . , θnT ) > 0.

So far, we know the information structure, we know what allocations are and we know what
allocations are feasible. But how does this guide towards a set of tax policies? In other
words the remaining question is, given the private information, what allocations can be
implemented?

One period economy

Suppose for now T = 1.

Definition 4 A Game (or a Mechanism) is a set of actions (A1 , . . . , AN ) and outcome

functions
N
Y
c y
(g , g ) : An −→ F A
n=1

The timing is as follows:

• Nature makes a draw θ for each agent n

• Agents privately observe θ and then simultaneously choose an action an ∈ An

31
ECN 741: Public Economics Fall 2008

• The outcome is determined according to outcome function

Definition 5 Let (A, g c , g y ) be a Mechanism. A Bayesian Nash Equilibrium (BNE) is a

collection of strategies {αn∗ }N ∗
n=1 , αn : Θ −→ An such that

∗
  y 
X gn (ρ, α−n (θ−n ))
αn∗ (θn ) ∈ arg max c ∗
π(θ−n ) u(gn (ρ, α−n (θ−n ))) − v
ρ∈An
θ−n
θn

We call gnc (α1∗ (θ1 ), . . . , αN

∗
(θN )) and gny (α1∗ (θ1 ), . . . , αN
∗
(θN )) ∀ (θ1 , . . . , θN ) equilibrium out-
come.

Definition 6 A feasible allocation (cn , yn )N

n=1 is implementable if there is a mechanism
c y ∗ N
(A, g , g ) and a BNE {αn }n=1 of that mechanism such that

cn = gnc (α1∗ (θ1 ), . . . , αN

∗
(θN )), yn = gny (α1∗ (θ1 ), . . . , αN
∗
(θN ))

So far we have made it clear what exactly do we mean by implementability. But is it helpful?
Notice that our setup so far does not impose any restriction on the type of games (mecha-
nisms) considered. Any equilibrium outcome of some game is implementable. Think for a
moment about the following problem: we want to find the best implementable allocation.
That means we need to search in the space of games, find a game that has a BNE that
implements that best allocation as its outcome. This is a very complicated problem.
Good news is that there is a very powerful result that allows us to restrict attention to a
very particular game without lose of generality. We need couple of more definitions.

Definition 7 A Direct Mechanism is a a game such that An = Θ for all n.

Definition 8 A truth-telling BNE of a direct mechanism (Θ, g c , g y ) is αn∗ (θn ) = θn for all n
such that   y 
X
c gn (ρ, θ−n ))
θn ∈ arg max π(θ−n ) u(gn (ρ, θ−n )) − v
ρ∈Θ
θ
θn
−n

A feasible allocation is truthfully implementable if

cn = gnc (θ1 , . . . , θN ), yn = gny (θ1 , . . . , θN )

In a direct mechanism, players are basically asked to report their skill type. We are interested
in equilibria in which type is revealed truthfully. It turns out there is not loss of generality
in doing that.

32
ECN 741: Public Economics Fall 2008

Proposition 9 (Revelation Principle)

A allocation (cn , yn )N
n=1 is implementable if and only if it is truthfully implementable in a
direct mechanism.

Proof.
Suppose allocation (cn , yn )N c y
n=1 is implementable as outcome of some mechanism (A, g , g ).
We construct a truth-full mechanism (Θ, g̃ c , g̃ y ) as the following

∗
g̃ c (θ1 , . . . , θN ) = g c (α1∗ (θ1 ), . . . , αN (θN )), ∗
g̃ y (θ1 , . . . , θN ) = g y (α1∗ (θ1 ), . . . , αN (θN ))

In which {αn∗ }N c y
n=1 is the BNE of the mechanism (A, g , g ). We only need to show that
truth-telling is a BNE of (Θ, g̃ c , g̃ y ). Suppose not, i.e., suppose there is a type θn and a
report ρ ∈ Θ such that
  y 
X
c g̃n (ρ, θ−n ))
π(θ−n ) u(g̃n (ρ, θ−n )) − v >
θ−n
θ n
  y 
X
c g̃n (θn , θ−n ))
π(θ−n ) u(g̃n (θn , θ−n )) − v =
θ−n
θn
  y ∗ ∗

X
c ∗ ∗ gn (αn (θn ), α−n (θ−n ))
π(θ−n ) u(gn (αn (θn ), α−n (θ−n ))) − v
θ
θn
−n

in which the last inequality follows from definition of (g̃ c , g̃ y ). This implies that there must
exist α = αn∗−1 (ρ) ∈ An such that
∗
gny (α, α−n
  
X
∗ (θ−n ))
π(θ−n ) u(gnc (α, α−n (θ−n ))) −v >
θ−n
θn
  y ∗ ∗

X
c ∗ ∗ gn (αn (θn ), α−n (θ−n ))
π(θ−n ) u(gn (αn (θn ), α−n (θ−n ))) − v
θ−n
θn

this is a contradiction. Therefore, (cn , yn )N

n=1 can be implemented by

cn = g̃ c (θ1 , . . . , θN ), yn = g̃ y (θ1 , . . . , θN )

Using Revelation Principle we can restrict attention to direct mechanism and allocations
that are truthfully revealing. This means that the set of implementable allocations are the

33
ECN 741: Public Economics Fall 2008

the ones that satisfy the following incentive compatibility constraints

yn (θ0 , θ−n )
   X   
X yn (θn , θ−n ) 0
π(θ−n ) u(cn (θn , θ−n )) + v ≥ π(θ−n ) u(cn (θ , θ−n )) + v
θ−n
θn θ
θn
−n

∀n, θn , θ0 ∈ Θ.
We are going to primarily focus on environment with unit measure of agents.

Environment with infinite number of agents

Consider the same environment as before (for general T < ∞) except that now there are
unit mass of agents. Nature makes a draw θT = (θ1 , . . . , θt ) ∈ ΘT = Θ × · · · × Θ for each
agent. The θT draws are i.i.d across agents. Let π(·) is probability density function over ΘT
draws. There is no aggregate uncertainty, therefore π(θT ) is also the mass of people who

have the draw θT . Let D ≡ θT |π(θT ) > 0 .
Define allocation as θt − measurable functions

ct : D −→ RT+

yt : D −→ RT+

Allocation is feasible if
T
XX T
XX
−t T
R ct (θ )π(θ ) ≤ T
R−t yt (θT )π(θT )
Θ∈D t=1 Θ∈D t=1

QT
Define a mechanism as set of actions A ⊂ t=1 Xt and outcome functions

g : A × ∆(A) −→ R2T
+

and gt (a, µ) = gt (a0 , µ0 ) if at = a0t and

X X
µ(āt , at+1 , . . . , aT ) = µ0 (āt , a0t+1 , . . . , a0T ) ∀āt
(at+1 ,...,aT ) (a0t+1 ,...,a0T )

in which µ, µ0 ∈ ∆(A) are measure of actions chosen.

34
ECN 741: Public Economics Fall 2008

A BNE is a strategy α∗ : D −→ A (αt is θt − measurable) such that

T  c ∗ T
gt (αt (θ ), µ∗ )
X  
t−1 c ∗ T
T ∗
β π(θ ) u(gt (αt (θ ), µ )) − v ≥
t=1
θt
T  c 0 T
gt (αt (θ ), µ∗ )
X  
t−1 T c 0 T ∗
β π(θ ) u(gt (αt (θ ), µ )) − v
t=1
θt

for all α0 : D −→ A (αt0 is θt − measurable) and µ∗ (a) = {θT |α(θT )=a} π(θT ). An allocation
P

is implementable if there exist a mechanism (A, g c , g y ) and a BNE α∗ such that

(ct (θT ), yt (θT )) = gt (α∗ (θT ), µ∗ )

A mechanism is a direct mechanism if A = ΘT . A strategy α is truth-full if αt (θT ) = θt

for all θT ∈ ΘT . An allocation is truthfully implementable if there exists a BNE of a direct
mechanism, i.e., if

T
gty (θT , µ∗ )
X   
β t−1 T
π(θ ) u(gtc (θT , µ∗ )) −v ≥
t=1
θt
T
gty (αt0 (θT ), µ∗ )
X   
β t−1
π(θ ) T
u(gtc (αt0 (θT ), µ∗ )) −v
t=1
θt

for all α0 : D −→ ΘT (αt0 is θt − measurable) µ∗ (a) = π(θT ) = π(a). And

P
{θT |θT =a}

(ct (θT ), yt (θT )) = gt (θT , π)

Revelation Principle: An allocation (c, y) is implementable only if it is truthfully imple-

mentable.
Note then that implementable allocation can be characterized by the following incentive
compatibility constraints

T X T X
yt (θT ) yt (αt0 (θT ))
X    X   
T T T 0 T
π(θ ) u(ct (θ )) − v ≥ π(θ ) u(ct (αt (θ ))) − v
t=1 θT ∈D
θt t=1 θT ∈D
θt

for all α0 : D −→ D (αt0 is θt − measurable).

35
ECN 741: Public Economics Fall 2008

Example : one period problem

Suppose T = 1 and there are only two types, θH and θL with θH > θL . Consider the
utilitarian planer’s problem

     
y(θH ) y(θL )
max π(θH ) u(c(θH )) − v + π(θL ) u(c(θL )) − v
θH θL
sub. to.
  
y(θH ) y(θL )
u(c(θH )) − v ≥ u(c(θL )) − v
θH θ
   H 
y(θL ) y(θH )
u(c(θL )) − v ≥ u(c(θH )) − v
θL θL

π(θH ) [c(θH ) − y(θH )] + π(θH ) [c(θH ) − y(θH )] = 0

For a moment suppose there is no private information. Then the optimal allocation must
satisfy (note that v(·) is convex):

u0 (c(θH )) = u0 (c(θL )) = λ ⇒ c(θH ) = c(θL )

   
1 y(θH ) 1 0 y(θL )
v = v = λ ⇒ y(θH ) > y(θL )
θH θH θL θL

Note that there is no distortion

 
1 y(θ)
u (c(θ)) = v 0
0
θ θ

We will show that this allocation does not satisfy I.C. constraints. Note that y(θH ) > y(θL ),
therefore

     
y(θL ) y(θL ) y(θH )
u(c(θL )) − v = u(c(θH )) − v > u(c(θH )) − v
θH θH θH
the I.C. for type H is violated.
So we know that when individuals have private information about their type, at least of the
the I.C. constraints is binding at the optimal solution.

36
ECN 741: Public Economics Fall 2008

Consider a relaxed planning problem with only type H’s I.C. constraint
     
y(θH ) y(θL )
max π(θH ) u(c(θH )) − v + π(θL ) u(c(θL )) − v
θH θL

sub. to.
   
y(θH ) y(θL )
u(c(θH )) − v ≥ u(c(θL )) − v ; π(θH )µ
θH θH

π(θH ) [c(θH ) − y(θH )] + π(θH ) [c(θH ) − y(θH )] = 0 ; λ

we will characterize the solution to this problem and then we will verify that at the solution
the I.C. constraint of type L is slack.

(1 + µ)u0 (c(θH )) = λ
 
π(θH )
1−µ u0 (c(θL )) = λ
π(θL )

 
1 0 y(θH )
(1 + µ) v = λ
θH θH
   
1 0 y(θL ) π(θH ) 1 0 y(θL )
v −µ v = λ
θL θL π(θL ) θH θH

First, note that there is no distortion for type H

 
0 1 0 y(θH )
u (c(θH )) = v
θH θH

Also, observe that

c(θH ) > c(θL )

note that incentive compatibility implies

   
y(θH ) y(θL )
v −v = u(c(θH )) − u(c(θL )) > 0
θH θH

and therefore
y(θH ) > y(θL )

37
ECN 741: Public Economics Fall 2008

Next we check that under these allocations, the I.C. constraints for type L is slack.

    Z y(θH )  Z y(θH )  
y(θH ) y(θL ) y y
v −v = v dy > v dy
θL θL y(θL ) θL y(θL ) θH
   
y(θH ) y(θL )
= v −v
θH θH
= u(c(θH )) − u(c(θL ))

rearrange terms    
y(θL ) y(θH )
u(c(θL )) − v > u(c(θH )) − v
θL θL

So, we know that our characterization is valid. Note that under this characterization

     
1 0 y(θL ) 0 π(θH ) 1 0 y(θL ) 0
v − u (c(θL )) = µ v − u (c(θL ))
θL θL π(θL ) θH θH
   
π(θH ) 1 0 y(θH ) 0
< µ v − u (c(θH ))
π(θL ) θH θH
< 0

and hence  
1 0 y(θL )
v < u0 (c(θL ))
θL θL
there is distortion for the low type.
Suppose we want to implement this allocation with a nonlinear tax function T (y). Let
T (y) = y − c if y ∈ {y(θH ), y(θL )} and T (y) = y otherwise. Consider the consumer’s
problem

y 
max u(c) − v
θ
sub. to.
c = y − T (y)

FOC
1 y 
u0 (c)(1 − T 0 (y)) = v
θ θ

38
ECN 741: Public Economics Fall 2008

discussion above implies that

T 0 (y(θL )) > 0, T 0 (y(θH )) = 0

3.1 The New Dynamic Public Finance

So far we have characterized the set of achievable allocations by any mechanism. The goal
of the planner is to find the best achievable allocation.

T
yt (θT )
X X   
t−1 T T
β π(θ )ω(θ1 ) u(ct (θ )) − v (43)
t=1
θt
θT ∈D

sub. to.
T X T X
yt (θT ) yt (αt0 (θT ))
X    X   
T T T 0 T
π(θ ) u(ct (θ )) − v ≥ π(θ ) u(ct (αt (θ ))) − v
t=1 θT
θt t=1 T
θt
θ

for all α0 : D −→ D (αt0 is θt − measurable).

T
XX T
XX
ct (θT )π(θT )/Rt−1 ≤ yt (θT )π(θT )/Rt−1
Θ∈D t=1 Θ∈D t=1

(ct , yt ) are θT -measurable

ct (θT ), yt (θT ) ≥ 0 ∀t, θT ∈ D

Note that we allow for allocation to depend on date on realization of θ.

The goal of this section is to characterize the properties of constraint efficient allocation (i.e.
the solution to the above planning problem). In particular we are interested in identifying
the optimal inter-temporal distortions.
To start we look at the full information problem.

Full information optima

Suppose θt is public information. Then the planning problem is the same, except that there
will no incentive compatibility constraint. Let λ be multiplier on feasibility.

X X
π(θT )ω(θ1 )u0 (ct (θT ))β t−1 = λ/Rt−1 π(θT )
θT θT

39
ECN 741: Public Economics Fall 2008

Note that this is the FOC with respect to a ct (θT ) at a particular draw θT . But we know
that ct (θT ) is θt − measurable, therefore we don’t need to sum over all θT ∈ D, but only
those that contain the particular history θt

X X
π(θT )ω(θ1 )u0 (ct (θT ))β t−1 = λ/Rt−1 π(θT )
θT |θt θT |θt

by measurability of ct (θT ).
u0 (ct (θT ))β t−1 ω(θ1 ) = λ/Rt−1

Note: This implies that optimal ct (θT ) is actually θ1 -measurable, i.e., it is independent from
θt for t > 1. In other words there is full insurance.
Note: The following Euler equation hold

u0 (ct (θT )) = βRE u0 (ct+1 (θT ))|θt

 

planner is happy to allow access to outside trade.

Note: Another Euler equation also holds.

1
= λ−1 β t−1 Rt−1 ω(θ1 )
u0 (c T
t (θ ))
 
1 1 t
= E 0 |θ
βR u (ct+1 (θT ))

3.1.1 Inverse Euler Equation

Consider again the original planning problem (43)(with incentive constraints). Let (c∗ , y ∗ )
be the solution to this problem. Now consider the following perturbation around (c∗ , y ∗ )

y0 = y∗

c0s = c∗s for all s 6= t, t + 1(for fixed t)

for all histories θt

u(c0t (θT )) + βu(c0t+1 (θT )) = k + u(c∗t (θT )) + βu(c∗t+1 (θT )) for all θt+1 such that π(θt+1 |θt ) > 0

40
ECN 741: Public Economics Fall 2008

X  X
π(θT ) c0t (θT ) + c0t+1 (θT )/R = π(θT ) c∗t (θT ) + c∗t+1 (θT )/R
  

θT |θt θT |θt

Note: (c0 , y 0 ) is feasible and incentive compatible.

Note: What we are doing is perturbing u(ct (θT )) by some amount and then make an appro-
priate perturbation in every immediate history following θt so that incentive compatibility is
preserved. If (c∗ , y ∗ ) is the solution to (43), this perturbation cannot improve welfare. One
implication of this is that (c∗ , y ∗ ) solves the following maximization problem and k = 0 at
the optimal solution.

max k
k,c0t (θT ),c0t+1 (θT )

sub. to

u(c0t (θT )) + βu(c0t+1 (θT )) = k + u(c∗t (θT )) + βu(c∗t+1 (θT )) for all θt , θt+1
such that π(θt+1 |θt ) > 0
X  X
π(θT ) c0t (θT ) + c0t+1 (θT )/R = π(θT ) c∗t (θT ) + c∗t+1 (θT )/R
  

θT |θt θT |θt

let η(θt+1 ) and λ be multipliers.

Let’s write the FOC
X X
η(θt+1 )u0 (c0t (θT )) = λ π(θT ) = λπ(θt )
θt+1 |θt θT |θt

for all θt , θt+1 such that π(θt+1 |θt ) > 0.1

X
βu0 (c0t (θt+1 ))η(θt+1 ) = λ/R π(θT ) = λπ(θt+1 )/R
θT |θt+1

Substitute for η(θt+1 )

X λπ(θt+1 )/R
0 (c0 (θ t+1 ))
u0 (c0t (θT )) = λπ(θt )
t+1 t
βu t
θ |θ

Cancel terms and evaluate this the solution c0 = c∗

βR X π(θt+1 ) 1
∗
= ∗
(44)
u0 (ct (θT )) t+1 t
π(θt ) u0 (ct+1 (θT ))
θ |θ

1
Note that I am abusing notation here. π(θt ) means the probability of history θt .

41
ECN 741: Public Economics Fall 2008

Note: The Intertemporal condition only depends on consumption which is observable.

Note: The additive separability assumption was key in deriving this result. Look at Farhi
and Werning (2008) for a version of this result that is derived for more general class of utility
functions.
Note: This result does not hold if private information affect the marginal utility of consump-
tion (for example in Atkeson and Lucas (1992) taste shock model).
This result implies that it is not desirable for planer to allow access to saving. To see this
look at the following Euler equation (which has to hold if there is access to saving)

X π(θt+1 )
u0 (ct (θT )) = βR t)
u0 (ct+1 (θT )) (45)
t+1 t
π(θ
θ |θ

But let’s look back at Inverse Euler Equation (44)

1
u0 (ct (θT )) = βR P π(θt+1 ) 1
θt+1 |θt π(θt ) u0 (c∗t+1 (θT ))
1
> βR 1
π(θ t+1 ) 0 ∗
u (ct+1 (θT ))
P
θ t+1 |θ t π(θ t )
X π(θt+1 )
= βR t
u0 (c∗t+1 (θT ))
t+1 t
π(θ )
θ |θ

Note that at the efficient allocation the individuals are “saving constrained”. In other words,
if individuals can privately save, they will choose to do so and it is desirable for planer to
prevent them from doing that.
Another way of seeing this is the following: suppose (45) holds. Then we must have

βR X π(θt+1 ) 1
0 ∗ T
< 0 ∗
u (ct (θ )) t+1 t
π(θ ) u (ct+1 (θT ))
t
θ |θ

Now suppose the planer wants to increase utility at time t by  and decrease it at time t + 1
by β −1 . The cost of increase of utility in period t is u0 (ct (θT ))/ . On the other hand planer
hands in u0 (ct+1 (θT ))/ less at each θt+1 that follows θt . Therefore it can free up resources.

42
ECN 741: Public Economics Fall 2008

On dynamics of consumption

Consider again the full information optimal allocation

u0 (ct (θT ))β t−1 ω(θ1 ) = λ/Rt−1

Suppose for simplicity βR = 1. Then

1. Allocation is independent of history (expect possibility θ1 )

2. There is no mobility in short-run or long-run

3. Inequality is constant

Now consider private information optimal allocations. Assume θt is i.i.d. Consider two
different history θt and θ̄t

X π(θt+1 )
u0 (ct (θT |θt )) = βR t)
u0 (ct+1 (θT |θt ))
t+1 t
π(θ
θ |θ

X π(θt+1 )
u0 (ct (θT |θ̄t )) = βR t)
u0 (ct+1 (θT |θ̄t ))
t+1 t
π(θ
θ |θ

note that π(θT |θ̄t ) = π(θT |θt ). Now suppose u0 (ct (θT |θt )) > u0 (ct (θT |θ̄t )), then there exist a
history θt+1 such that π(θt+1 |θt ) = π(θt+1 |θ̄t ) and

u0 (ct+1 (θT |θt )) > u0 (ct+1 (θT |θ̄t ))

good shocks up to period t has persistent effect on period t + 1 allocations.

Next we consider inequality. Assume, u(c) = log(c), then u01(c) = c. Start from inverse Euler
equation (βR = 1)  
1 1 t
=E 0 |θ
u0 (ct (θT )) u (ct+1 (θT ))

We want to know what happens to variance of consumption over time

    
1 1 t
V ar 0
= V ar E 0 |θ
u (ct (θT )) u (ct+1 (θT ))
    
1 1 t
= V ar − E V ar |θ
u0 (ct+1 (θT )) u0 (ct+1 (θT ))

43
ECN 741: Public Economics Fall 2008

 
1 t
If V ar u0 (ct+1 (θT ))
|θ > 0 for some θt , then

   
1 1
V ar 0
< V ar
u (ct (θT )) u0 (ct+1 (θT ))

and therefore
V ar ct (θT ) < V ar ct+1 (θT )

So inequality grows. And it is efficient.

How about mobility? In short-run there is mobility. What about long-run?
h i
1 1
Note that u0 (ct )
is a martingale (βR = 1). We also know that (by feasibility) E u0 (ct+1 )
<∞
Martingale Convergence Theorem: If {xt }∞ t=1 is stochastic process adapted to filteration
∞
{Ft }t=1 such that xt = E [xt+1 |Ft ] and E [xt ] < ∞ for all t, then

a.s.
lim xt = x∞ < ∞
t→∞

where x∞ is a random variable with E[x∞ ] < ∞.

1
Therefore, u0 (ct )
converges to a finite number and hence there is no mobility in the long-run.

3.1.2 Long-run properties of efficient allocations

This part is mostly based on Farhi and Werning (2006, 2007, 2005), PHELAN (2006) and
Atkeson and Lucas (1992).
In what follows we maintain the following assumptions:

• T =∞

• Θ = {θH , θL }

• θt i.i.d over time

Very important note: In what follows I skip some detailed steps. Most of the arguments
are heuristic and have loose ends. For more rigorous proofs please look at the references
above.

44
ECN 741: Public Economics Fall 2008

Immiseration result

For this part assume βR = 1. Consider the following planning problem

T
yt (θt )
X X   
t−1 t t
w0 = max β π(θ ) u(ct (θ )) − v (46)
t=1
θt
θT ∈D

sub. to.
T X T X
yt (θt ) yt (αt0 (θt ))
X    X   
t t
π(θ ) u(ct (θ )) − v ≥ t
π(θ ) u(ct (αt0 (θt ))) −v
θt θt
t=1 θt t=1 θt

for all α0 : D −→ D (αt0 is θt − measurable).

T
XX
π(θt ) ct (θt ) − yt (θt ) /Rt−1 ≤ 0
 

θt t=1

The solution to this problem must also be the solution to the following dual problem

T
XX
π(θt ) ct (θt ) − yt (θt ) /Rt−1
 
K(w0 ) = min (47)
θt t=1

sub. to
T X T X
yt (θt ) yt (αt0 (θt ))
X    X   
t t
π(θ ) u(ct (θ )) − v ≥ t
π(θ ) u(ct (αt0 (θt ))) −v
θt θt
t=1 θt t=1 θt

T
yt (θt )
X X   
t−1 t t
β π(θ ) u(ct (θ )) − v ≥ w0
t=1
θt
θT ∈D

In which K(U0 ) is the cost of delivering ex-ante utility U0 to everyone.

We want to write this recursively. Consider an allocation sequence (ct (θt ), yt (θt )). Consider
a history θ̄t . Then define the ex ante utility of an agent with history θ̄t under this plan as
∞ X
yt (θs )
X   
t s s−t s
wt (θ̄ ) = π(θ )β u(cs (θ )) − v
s=t θs |θ̄t
θs

we call wt (θ̄t ) the “promised utility” after history θ̄t .

Let (c∗t (θt ), yt∗ (θt )) be the solution to problem (47) wt∗ (θ̄t ) be promised utility after history θ̄t .
We can show that (c∗t (θ̄t , θt+1 ), yt∗ (θ̄t , θt+1 ), wt∗ (θ̄t , θt+1 )) solve the following Bellman Equation

45
ECN 741: Public Economics Fall 2008

at w = wt∗ (θ̄t ) (note: I imposed the βR = 1 assumption here)

X
K(w) = min π(θ) [c(θ, w) − y(θ, w) + βK(w0 (θ, w))]
c,y,W
θ

sub. to

y(θ0 , w)
   
y(θ, w) 0 0
u(c(θ, w)) − v + βw (θ, w) ≥ u(c(θ , w)) − v + βw0 (θ0 , w) ∀θ, θ0
θ θ
   
X y(θ, w) 0
π(θ) u(c(θ, w)) − v + βw (θ, w) ≥ w
θ
θ

The second constraint is called “promise keeping” constraint.

Proposition 10 K(w) is strictly increasing and strictly convex (assumptions on v(·) is

needed). Also. Let w and w̄ be the lowest and highest possible values for promised utility.
Then, limw→w K 0 (w) = 0 and limw→w̄ K 0 (w) = limw→w̄ K(w) = 0.

Let µ(θ, θ0 ) be multiplier on incentive constraint and φ the multiplier on promise keeping.
First order condition with respect to c(θ, U ) is
 
X X
u0 (c(θ, w))  µ(θ, θ0 ) − µ(θ0 , θ) + π(θ)φ = π(θ)
θ0 θ0

and with respect to w0 (θ, U )

 
X X
 µ(θ, θ0 ) − µ(θ0 , θ) + π(θ)φ = π(θ)K 0 (w0 (θ, w)) (48)
0 0
θ θ

and therefore
1
K 0 (w0 (θ, w)) =
u0 (c(θ, w))

this implies the following lemma

Lemma 1 Given any w ∈ [w, w̄], if w0 (θ, w) = w0 (θ0 , w) for some θ, θ0 ∈ Θ, then c(θ, w) =
c0 (θ, w)

Sum the equation (48) over all θ

X XX XX
π(θ)K 0 (w0 (θ, w)) = µ(θ, θ0 ) − µ(θ0 , θ) + φ = φ
θ θ θ0 θ θ0

46
ECN 741: Public Economics Fall 2008

also from envelope condition we have

K 0 (w) = φ

therefore
X
K 0 (w) = π(θ)K 0 (w0 (θ, w))
θ

Start from a given w0 , construct a stochastic process wt as

wt+1 = w0 (θt , wt )

then
K 0 (wt ) = Et [K 0 (wt+1 )]

hence wt is a martingale. By martingale convergence theorem there must exist a w∞ such

0
that wt a.s.
−→w∞ . Suppose K (w∞ ) > 0. Note that convergence implies that

w0 (θ, w∞ ) = w0 (θ0 , w∞ ) ∀θ, θ0

and therefore
c(θ, w∞ ) = c(θ0 , w∞ ) ∀θ, θ0

and then incentive compatibility implies

y(θ, w∞ ) = y(θ0 , w∞ ) ∀θ, θ0

but we know from our two type static example that the planer can do better by differentiating
various θ types. Therefore, this is a contradiction. Hence K 0 (w∞ ) = 0 and w∞ = w.

No Immiseration result

Consider the following planning problem in which planer values future consumption more
than agent (β̂ > β)

T
yt (θt )
X X   
t−1 t t
w0 = max β̂ π(θ ) u(ct (θ )) − v̂ (49)
t=1
θt
θT ∈D

47
ECN 741: Public Economics Fall 2008

sub. to.
T X T X
yt (θt ) yt (αt0 (θt ))
X    X   
t t
π(θ ) u(ct (θ )) − v̂ ≥ t
π(θ ) u(ct (αt0 (θt ))) − v̂
θt θt
t=1 θt t=1 θt

for all α0 : D −→ D (αt0 is θt − measurable).

T
XX
π(θt ) ct (θt ) − yt (θt ) /Rt−1 ≤ 0
 

θt t=1

assume the following (in addition to the assumption mentioned at the beginning of the
discussion)

• β̂R = 1
y

v(y)
• v̂ θ
= θ

• u() is unbounded below, hence w = −∞

1
• E θ
= 1.

Suppose the above problem has a solution and let λ̂be the multiplier on resources constraint.
Then the solution to the above problem must also solve the following

T
v(yt (θt ))
X X  
t−1 t t t t
P (w0 ) = max β̂ π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ ) (50)
t=1
θt
θT ∈D

sub. to.
T X T X
v(yt (θt )) v(yt (αt0 (θt )))
X   X  
t t t 0 t
π(θ ) u(ct (θ )) − ≥ π(θ ) u(ct (αt (θ ))) −
θt θt
t=1 θt t=1 t θ

for all α0 : D −→ D (αt0 is θt − measurable).

T
v(yt (θt ))
X X  
t−1 t t
β π(θ ) u(ct (θ )) − ≥ w0
t=1
θt
θT ∈D

Again, we can show that the solution to the above problem also solves the following Bellman
equation (after any history)

48
ECN 741: Public Economics Fall 2008

 
X v(y(θ, w)) 0
P (w) = max0 π(θ) u(c(θ, w)) − − λ̂c(θ, w) + λ̂y(θ, w) + β̂P (w (θ, w)) (51)
c,y,w
θ
θ

sub. to

v(y(θ, w)) v(y(θ0 , w))

u(c(θ, w)) − + βw0 (θ, w) ≥ u(c(θ0 , w)) − + βw0 (θ0 , w) ∀θ, θ0
θ θ
 
X v(y(θ, w)) 0
π(θ) u(c(θ, w)) − + βw (θ, w) ≥ w
θ
θ

We want to show that in this problem in the long-run the promised utility cannot be at
misery (w∞ > −∞ ). We use two lemmas to show this

Lemma 2 The value function P (w) is strictly concave and continuously differentiable on
(−∞, w̄). Moreover,

lim P (w) = lim P (w) = lim P 0 (w) = −∞

v→−∞ v→w̄ v→w̄

and
lim P 0 (w) = 1
v→−∞

Proof.
We are not going to prove concavity and differentiability. We take them as given.
Warning: This proof is not complete! There are some steps that needs to be filled in or
reformulated. I present it to provide the core idea of the proof as I understand it.
Warning:The proof that I presented in class is wrong! The mistake is in constructing the
upper and lower bound value function (PF I and Pm ). In particular, I mentioned in class
that Pi (w) = w − Ki (w) (for i = F I, m, etc), in which Ki (w) is the cost of delivering utility
w. This is correct only if β̂ = β. When β̂ > β individual utilities are discounted at different
rate in the planner’s objective and expected present discounted value of an allocation in the
planner’s objective is not the same as the one promised to the agent.
Correct proof has the same idea but those functions are constructed correctly.
Define value function PF I (w) as

T
v(yt (θt ))
X X  
t−1 t t t t
PF I (w) = max β̂ π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1
θt
θT ∈D

49
ECN 741: Public Economics Fall 2008

sub. to.
T
v(yt (θt ))
X X  
t−1 t t
β π(θ ) u(ct (θ )) − ≥w
t=1
θt
θT ∈D

PF I (w) is the value to the planner from delivering utility w to individual, if we ignore
incentive constraint. Note that PF I (w) > P (w). Also, PF I (w) is strictly concave and
differentiable.
Next consider the following maximization problem

v(y)
m = max u(c) − − λ̂c + λ̂y
c,y,θ θ

The above problem has a solution (u0 (c) = λ̂,v 0 (y) = λ̂θ, and θ belongs to a compact set).
Next, note that

T
v(yt (θt ))
X X  
t−1 t t t t
PF I (w) = β̂ π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1 T
θt
θ ∈D
T
v(yt (θt ))
X X  
t−1 t t t t
= β̂ π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1
θ t
θT ∈D
T
v(yt (θt ))
X X  
t−1 t t t t
+ β π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1 T
θt
θ ∈D
T
v(yt (θt ))
X X  
t−1 t t t t
− β π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1 T
θt
θ ∈D
T
X X h i
= w+ β t−1 π(θt ) −λ̂ct (θt ) + λ̂yt (θt ) +
t=1 θT ∈D
T
v(yt (θt ))
X X  
t−1 t−1 t t t t
(β̂ − β ) π(θ ) u(ct (θ )) − − λ̂ct (θ ) + λ̂yt (θ )
t=1
θt
θT ∈D
T  
X
t−1
X
t
h
t t
i 1 1
≤ w+ β π(θ ) −λ̂ct (θ ) + λ̂yt (θ ) + m −
t=1 T 1 − β̂ 1−β
θ ∈D
 
1 1
≤ w − λ̂K̃(w) + m −
1 − β̂ 1 − β

in which
T
X X
β t−1 π(θt ) ct (θt ) − yt (θt )
 
K̃(w) = min
t=1 θT ∈D

50
ECN 741: Public Economics Fall 2008

sub. to.
T
v(yt (θt ))
X X  
t−1 t t
β π(θ ) u(ct (θ )) − ≥w
t=1
θt
θT ∈D

Note that K̃(w) is strictly convex and differentiable and limw→−∞ K̃ 0 (w) = 0.
Let Pmax (w) = w − K̃(w) + m. Then Pmax (w) ≥ PF I (w) and both are strictly con-
cave. Also, limw→−∞ PF I (w) ≤ limw→−∞ Pmax (w) = −∞. Therefore, limw→−∞ PF0 I (w) ≤
0
limw→−∞ Pmax (w) = 1.
Also, Note that limw→−∞ P (w) ≤ limw→−∞ PF I (w) = −∞ and therefore (since both are
strictly concave)
lim P 0 (w) ≤ lim PF0 I (w) = 1.
w→−∞ w→−∞

lim P 0 (w) ≤ 1
w→−∞

Next, consider allocations (c(w0 , θt ), y(w0 , θt )) that solve the original problem. Suppose they
attain the value P (w0 ). Define new allocations (c̃(w, θt ), ỹ(w, θt )) for w ≤ w0 as

c̃(w, θt ) = c(w0 , θt ) ∀θt ∀t

ỹ(w, θt ) = y(w0 , θt ) ∀θt ∀t > 1
ỹ(w, θ1 ) = v −1 (v(y(w0 , θ1 )) + w0 − w)

Now define Pm (w) for w ≤ w0 as

T
v(ỹ(w, θt ))
X X  
t−1 t t t t
Pm (w) = β̂ π(θ ) u(c̃(w, θ )) − − λ̂c̃(w, θ ) + λ̂ỹ(w, θ )
t=1
θt
θT ∈D
 
X v(y(w0 , θ1 )) + w0 − w
= π(θ1 ) u(c(w0 , θ1 )) − − λ̂c(w0 , θ1 ) + λ̂ỹ(w0 , θ1 ) +
θ1
θ1

T
v(c(w0 , θt ))
X X  
t−1 t t t t
β̂ π(θ ) u(c(w0 , θ )) − − λ̂c(w0 , θ ) + λ̂y(w0 , θ )
t=2
θt
θT ∈D
  X  
X w − w0 v(y(w0 , θ1 ))
= π(θ1 ) + λ̂ỹ(w, θ1 ) + π(θ1 ) u(c(w0 , θ1 )) − − λ̂c(w0 , θ1 )
θ1
θ1
θ1
θ1

T
v(c(w0 , θt ))
X X  
t−1 t t t t
+ β̂ π(θ ) u(c(w0 , θ )) − − λ̂c(w0 , θ ) + λ̂y(w0 , θ )
t=2 T
θ t
θ ∈D

51
ECN 741: Public Economics Fall 2008

Note that Pm (w) is strictly concave, Pm (w) ≤ P (w). Also, note that only the first term
depends on w. Therefore
 
1
Pm0 (w) = 1 − λ̂E 0
v (v(y(w, θ0 )) + w0 − w)

and limw→−∞ Pm0 (w) = 1.

Therefore, limw→−∞ P 0 (w) ≥ limw→−∞ Pm0 (w) = 1. Hence, we proved that limw→−∞ P 0 (w) =
1.

In the next lemma we show that 1 − P (w0 (w, θ)) can be bounded above and bellow for all θ.
Let’s rewrite the problem (51) again
 
X v(y(θ, w)) 0
P (w) = max0 π(θ) u(c(θ, w)) − − λ̂c(θ, w) + λ̂y(θ, w) + β̂P (w (θ, w)) (52)
c,y,w
θ
θ

sub. to

v(y(θ, w)) v(y(θ0 , w))

u(c(θ, w)) − + βw0 (θ, w) ≥ u(c(θ0 , w)) − + βw0 (θ0 , w) ∀θ, θ0
θ θ
 
X v(y(θ, w)) 0
π(θ) u(c(θ, w)) − + βw (θ, w) ≥ w
θ
θ

Suppose only high type’s incentive constraint binds. Let µ and φ be multipliers on the IC
and promise keeping (and let π(θH ) = π)
FOC w.r.t c(w, θ)
 
1
π 1 − λ̂ 0 + φπ + µ = 0
u (c(w, θH ))
 
1
(1 − π) 1 − λ̂ 0 + φπ − µ = 0
u (c(w, θL ))

FOC w.r.t w0 (w, θ)

π β̂P 0 (w, θH ) + φπβ + βµ = 0

(1 − π)β̂P 0 (w, θL ) + φ(1 − π)β − βµ = 0

52
ECN 741: Public Economics Fall 2008

FOC w.r.t y(w, θ)

π 1 π µ
− + π λ̂ 0 −φ − = 0
θH v (y(w, θH )) θH θH
(1 − π) 1 (1 − π) µ
− + (1 − π)λ̂ 0 −φ + = 0
θL v (y(w, θL )) θL θH

combining these FOC’s together with envelope condition P 0 (w) = −φ we get

β β
E [1 − P 0 (w0 (w, θ))] = (1 − P 0 (w)) + 1 −
β̂ β̂

and    
1 1
λ̂E 0 = (1 + φ)E =1+φ
v (y(w, θ)) θ

Assume that we know y(w, θH ) > y(w, θL ) and w0 (w, θH ) > w0 (w, θL ). Now we can prove
the next lemma.

Lemma 3 The following inequalities hold

 
0 β θH β β β
(1 − P (w)) 1+ − θL +1− ≤ 1 − P 0 (w0 (w, θ)) ≤ (1 − P 0 (w)) θH + 1 −
β̂ θL β̂ β̂ β̂

Proof.
From the FOC for y(w, θH ) we get
" #
λ̂ λ̂ µ
(1 + φ)θH = E 0 θH ≥ 0 =1+φ+
v (y(w, θ)) v (y(w, θ)) π

µ
from this we can get the following bound on π

µ
≤ (1 + φ)(θH − 1)
π

Now we use this in the FOC for w0 (w, θH ) and we get

β β
P 0 (w0 (w, θH )) ≥ −φ θH − (θH − 1)
β̂ β̂
β β
= P 0 (w) θH − (θH − 1)
β̂ β̂

53
ECN 741: Public Economics Fall 2008

after rearranging terms

β β
1 − P 0 (w0 (w, θL )) < 1 − P 0 (w0 (w, θH )) ≤ (1 − P 0 (w)) θH + 1 −
β̂ β̂

Using similar argument we can show that

µ θH
≤ (1 + φ)(1 − θL )
1−π θL

and
   
0 0 β θH β θH
P (w (w, θL ) ≤ −φ 1+ − θL + − θL
β̂ θL β̂ θL
   
0 β θH β θH
= P (w) 1+ − θL + − θL
β̂ θL β̂ θL

and hence

 
0 0 0 0 0 β θH β
1 − P (w (w, θH )) > 1 − P (w (w, θL )) ≥ (1 − P (w)) 1+ − θL + 1 −
β̂ θL β̂

Now let w → −∞, then P 0 (w) → 1 and

β
lim P 0 (w0 (w, θ)) = <1
w→−∞ β̂

So w0 (w, θ) can never stay at misery level.

For more detailed arguments and complete proof of the existence of stationary distribution
see Farhi and Werning (2006, 2007, 2005).

3.1.3 Aggregate Risk

In this section we derive inverse Euler equation in an environment in which there is aggregate
risk. We need to introduce new notations. Let Θ be the space of individual shocks and Z
be the space of aggregate shock. The timing is the following

1. Nature draws z T ∈ Z T according to p.d.f πz (z T ).

54
ECN 741: Public Economics Fall 2008

2. Nature draw individual shocks θT ∈ ΘT according to πθ (θT |z T ). These draws are i.i.d
across individuals conditional on z T .

By law of large number, given z T , the fraction of population with shocks θT is πθ (θT |z T ).
We impose the following restriction:

Assumption 1 For all θT ∈ ΘT , πθ (θt |z T ) = πθ (θt , θt+1 , . . . , θT |z T ) is indepen-

P
(θt+1 ,...,θT )
dent of zt+1 , . . . , zT .

This assumption implies that conditional on z t , (θt+1 , . . . , θT ) and (zt+1 , . . . , zT ) are inde-
pendent.
Remarks: Note that in this setup we are not imposing any restriction on time series prop-
erties of θt and zt . However, our assumption implies that by observing history of private
shocks up to date t the individuals cannot infer anything about future aggregate shocks.
As before, we assume that agents learn zt and θt at the beginning of date t and θt is private
information.
There is an initial capital stock K̄1 in the economy.

Definition 9 An allocation is a sequence of functions (c, y, K) such that

c : ΘT × Z T → RT+ , ct : (θt , z t )-measurable

y : ΘT × Z T → RT+ , yt : (θt , z t )-measurable

K : Z T → RT+ , Kt+1 : z t -measurable

Definition 10 An allocation is feasible if

Ct (z T ) + Kt+1 (z T ) = (1 − δ)Kt (z T ) + F (Kt (z T ), Yt (z T ), z T ) (53)

X
Ct (z T ) = πθ (θT |z T )ct (θT , z T )
θT ∈ΘT
X
Yt (z T ) = πθ (θT |z T )yt (θT , z T )
θT ∈ΘT
K1 = K̄1

55
ECN 741: Public Economics Fall 2008

Definition 11 An allocation is incentive compatible if

T
yt (θT , z T )
X X X   
T t−1 T T T T
πz (z ) β πθ (θ |z ) u(ct (θ , z )) − v ≥ (54)
t=1
θt
z T ∈Z T θT ∈ΘT
T
yt (αt0 (θT , z T ))
X X X   
T
πz (z ) β t−1 T
πθ (θ |z ) T
u(ct (αt0 (θT , z T ))) −v
t=1
θt
z T ∈Z T θT ∈ΘT

for all α0 : Θ −→ Θ (αt0 is (θt , z t ) − measurable).

The planing problem

T
yt (θT , z T )
X X X   
T t−1 T T T T
max πz (z ) β πθ (θ |z ) u(ct (θ , z )) − v
t=1
θt
z T ∈Z T θT ∈ΘT

sub. to (53) and (54).

The inverse Euler equation

Suppose (c∗ , y ∗ , K ∗ ) is the solution to the above problem. Fix a public history z̄ t . We perturb
the solution to (c0 , y ∗ , K 0 ) such that
X
u(c0t (θt , z̄ t )) = u(c∗t (θt , z̄ t )) + β πz (z̄ t , zt+1 )δt+1 (θt , z̄ t , zt+1 ) + γ ∀θt (55)
zt+1

u(c0t+1 (θt+1 , z̄ t , zt+1 )) = u(c∗t+1 (θt+1 , z̄ t , zt+1 )) − δt+1 (θt , z̄ t , zt+1 ) (56)

0
X X
π(θt |z̄ t )c0t (θt , z̄ t ) + Kt+1 (z̄ t ) ≤ π(θt |z̄ t )c∗t (θt , z̄ t ) + Kt+1
∗
(z̄ t ) (57)
θt θt

0 0 0
X
πθ (θt+1 |z̄ t , zt+1 )ct+1 (θt+1 , z̄ t , zt+1 ) − Kt+1 (z̄ t )(1 − δ) − F (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 )(58)
θt+1
≤
X
πθ (θt+1 |z̄ t , zt+1 )c∗t+1 (θt+1 , z̄ t , zt+1 ) − Kt+1
∗ ∗
(z̄ t )(1 − δ) − F (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 )
θt+1

The idea is that (c∗ , y ∗ , K ∗ ) must be the solution to the following maximization problem
(with γ = 0)
0= max 0 γ
0 0 δt+1 ,ct ,ct+1 ,γ,Kt+1

56
ECN 741: Public Economics Fall 2008

sub. to (55)-(58).
Let η(θt , z̄ t ),η(θt+1 , z̄ t , zt+1 ), µ(z̄ t ) and µ(z̄ t , zt+1 ) be multipliers on (55), (56), (57) and (58).
Write the FOCs

u0 (c∗t (θt , z̄ t ))η(θt , z̄ t ) = µ(z̄ t )πθ (θt |z̄ t ) (59)

u0 (c∗t+1 (θt+1 , z̄ t , zt+1 ))η(θt+1 , z̄ t , zt+1 ) = µ(z̄ t , zt+1 )πθ (θt+1 |z̄ t , zt+1 ) (60)
X
βπz (z̄ t , zt+1 )η(θt , z̄ t ) = η(θt+1 , z̄ t , zt+1 ) (61)
θt+1 ,zt+1 |θt ,z̄ t
X
∗
µ(z̄ t ) = µ(z̄ t , zt+1 )(1 − δ + FK (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 ))
(62)
zt+1

substitute η(θt ) and η(θt+1 , z̄ t , zt+1 ) from (59) and (60) into (61)

µ(z̄ t )πθ (θt |z̄ t ) X µ(z̄ t , zt+1 )πθ (θt+1 |z̄ t , zt+1 )
βπz (z̄ t , zt+1 ) =
u0 (c∗t (θt , z̄ t )) u0 (c∗t+1 (θt+1 , z̄ t , zt+1 ))
θt+1 ,zt+1 |θt ,z̄ t

Take µ(z̄ t , zt+1 ) out of summation and rearrange terms

 −1
t t t t+1 t
βπz (z̄ , zt+1 )πθ (θ |z̄ )  X πθ (θ |z̄ , zt+1 )  µ(z̄ t , zt+1 )
=
u0 (c∗t (θt , z̄ t )) u (c∗t+1 (θt+1 , z̄ t , zt+1 ))
0 µ(z̄ t )
θt+1 ,zt+1 |θt ,z̄ t

Note that by the Independence assumption we had

πθ (θt+1 |z̄ t , zt+1 ) πθ (θt+1 |z̄ t , zt+1 )

= = πθ (θt+1 |z̄ t , zt+1 , θt )
πθ (θt |z̄ t ) πθ (θt |z̄ t , zt+1 )

Let
 −1
t t+1 t
µ(z̄ , zt+1 ) β X πθ (θ |z̄ , zt+1 ) 
λ(z̄ t , zt+1 ) ≡ = 0 ∗ t t 
t t
µ(z̄ )πz (z̄ , zt+1 ) u (ct (θ , z̄ )) u (c∗t+1 (θt+1 , z̄ t , zt+1 ))
0
θt+1 |z̄ t ,zt+1 ,θt

then
X
∗
1= πz (z̄ t , zt+1 )λ(z̄ t , zt+1 )(1 − δ + FK (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 ))
zt+1

or

   −1
t β 1  t+1 t
λ(z̄ , zt+1 ) = 0 ∗ t t E  θ , z̄ , zt+1
u (ct (θ , z̄ )) u0 (c∗t+1 (θt+1 , z̄ t , zt+1 )) 

57
ECN 741: Public Economics Fall 2008

∗
1 = E λ(z̄ t , zt+1 )(1 − δ + FK (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 ))|z̄ t
 

Note that (using Jensen’s inequality)

   −1
t β 1  t+1 t
λ(z̄ , zt+1 ) = 0 ∗ t t E 0 ∗  θ , z̄ , zt+1
u (ct (θ , z̄ )) u (ct+1 (θt+1 , z̄ t , zt+1 )) 

βE u0 (c∗t+1 (θt+1 , z̄ t , zt+1 )) θt+1 , z̄ t , zt+1
 
<
u0 (c∗t (θt , z̄ t ))

and therefore

u0 (c∗t (θt , z̄ t )) < βE u0 (c∗t+1 (θt+1 , z̄ t , zt+1 ))(1 − δ + FK (Kt+1

∗
(z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 )) θt+1 , z̄ t , zt+1
  

Example 1: suppose Θ is singleton. Then

t βu0 (c∗t+1 (z̄ t , zt+1 ))

λ(z̄ , zt+1 ) =
u0 (c∗t (z̄ t ))
and therefore

βu0 (c∗t+1 (z̄ t , zt+1 ))

 
∗
1=E (1 − δ + FK (Kt+1 (z̄ t ), Yt∗ (z̄ t , zt+1 ), z̄ t , zt+1 ))|z̄ t
u0 (c∗t (z̄ t ))

Example 2: suppose Z is singleton. Then

 t+1 −1
   
β 1
λt+1 = 0 ∗ t E θ
u (ct (θ )) u0 (c∗t+1 (θt+1 )) 
∗
1 = λt+1 (1 − δ + FK (Kt+1 , Yt∗ ))

and therefore ∗
, Yt∗ ))

β(1 − δ + FK (Kt+1
 
1  t+1
=E 0 ∗ θ
u0 (c∗t (θt )) u (ct+1 (θt+1 )) 
and therefore (using Jensen’s inequality)

u0 (c∗t (θt )) < β(1 − δ + FK (Kt+1

∗
, Yt∗ ))E u0 (c∗t+1 (θt+1 )) θt+1
  

3.1.4 Inter-temporal optimality with balanced growth preferences

The additive separable preferences that we have assumed so far are not consistent with
balanced growth fact (except if u(c) = log(c)). However, the perturbation that we used

58
ECN 741: Public Economics Fall 2008

in deriving the inverse Euler equation depends crucially on additive separability. If the
preferences are of the form
c1−σ
u(c, l) = v(l)
1−σ
then the proof presented will not work. See Farhi and Werning (2008) for derivation of the
inter-temporal optimality condition for larger class of preferences (including those that are
consistent with balanced growth).

3.2 Implementing Efficient Allocations

3.2.1 Wedges and Taxes

So far we have shown that efficient allocations must satisfy a condition like the following (if
there is no aggregate shock)

∗
, Yt∗ ))

β(1 − δ + FK (Kt+1
 
1  t+1
=E 0 ∗  θ .
u0 (c∗t (θt )) u (ct+1 (θt+1 )) 
We also showed that this implies the following

u0 (c∗t (θt )) < βE u0 (c∗t+1 (θt+1 ))(1 − δ + FK (Kt+1

∗
, Yt∗ )) θt+1 .
  

In other words there is an inter-temporal wedge

u0 (c∗t (θt ))
1 − τt =    < 1.
βE u0 (c∗t+1 (θt+1 ))(1 − δ + FK (Kt+1
∗
, Yt∗ )) θt+1

But does this mean that the efficient allocations can be implemented by a positive tax on
capital?

Two period example

Consider the following example:

• T = 2.

• Θ = {0, 1}, π(θ1 = 1) = 1, π(θ2 = (1, 1)) = 0.5 and π(θ2 = (1, 0)) = 0.5. (Note that
this implies y2h = y2 ((1, 1)) = l2h (1, 1) and y2l = y2 ((1, 0)) = 0)
2
• u(c, l) = log(c) − l2 , β = 1.

59
ECN 741: Public Economics Fall 2008

• F (K, Y ) = RK + wY , δ = 1.

• There is endowment of K1 in period 1.

The planner’s problem is

y12 2
 
y2h
max log(c1 ) − + 0.5 log(c2h ) − + 0.5 [log(c2l )]
c1 ,c2h ,c2l ,y1 ,y2h ,K2 2 2

sub. to

c1 + K2 = RK1 + wy1
0.5c2h + 0.5c2l = RK2 + 0.5wy2h
2
y2h
log(c2h ) − ≥ log(c2l )
2
c1 , c2h , c2l , y1 , y2h , K2 ≥ 0

Let (c∗1 , c∗2h , c∗2l , y1∗ , y2h

∗
, K2∗ ) be the solution. Then it should satisfy the following FOC

c∗1 + K2∗ = RK1 + wy1∗

0.5c∗2h + 0.5c∗2l = RK2∗ + 0.5wy2h
∗

∗2
y2h
log(c∗2h ) − = log(c∗2l )
2
Rc∗1 = 0.5c∗2h + 0.5c∗2l

w ∗
= y2h
c∗2h
w
= y1∗
c∗1

We want to implement these allocations in a competitive equilibrium. We assume the fol-

lowing market structure

• There is a single firm that rents capital and labor.

• Individuals have the same endowment of k1 = K1 and choose how much to work and
how much to save.

60
ECN 741: Public Economics Fall 2008

• Government taxes savings at rate τk .

• Government makes transfers T2h and T2l to individual who produce y2h and zero output
in the second period. Transfers can be negative.

Competitive equilibrium is allocations (c1 , y1 , c2h , y2h , c2l , k2 ) and policy (τk , T2h , T2l ) such
that:

1. Given the policy, allocations solve consumer problem

y12 2
 
y2h
max log(c1 ) − + 0.5 log(c2h ) − + 0.5 [log(c2l )]
c1 ,c2h ,c2l ,y1 ,y2h ,k2 2 2

sub. to
c1 + k2 = Rk1 + wy1

c2h = R(1 − τk )k2 + wy2h + T2h , if y2h > 0

c2h = R(1 − τk )k2 + T2l , if y2h > 0
c2l = R(1 − τk )k2 + T2l

c1 , c2h , c2l , y1 , y2h , k2 ≥ 0

2. Markets clear

c1 + K2 = RK1 + wy1
0.5c2h + 0.5c2l = RK2 + 0.5wy2h
k1 = K1
k2 = K2

Note that the government budget constraint is satisfied in equilibrium

Rτk K2 = 0.5T2h + 0.5T2l

We want the tax system be such that y2h > 0. Then it is necessary that
2
y2h
log(c2h ) − ≥ log(c2l )
2

61
ECN 741: Public Economics Fall 2008

Let’s assume we have such a tax system. Then the consumer FOCs are
 
1 .5 .5
= R(1 − τk ) +
c1 c2h c2l

w
= y1
c1
w
= y2h
c2h

Also note that

c2h = R(1 − τk )k2 + wy2h + T2h

c2l = R(1 − τk )k2 + T2l

∗
Question: How do we pick taxes to implement efficient allocations (c∗1 , c∗2h , c∗2l , y1∗ , y2h , K2∗ )
in the equilibrium?
Why don’t we try the simplest way to do this. And choose the following taxes

[0.5c∗2h + 0.5c∗2l ]−1

1 − τk = h i
0.5 0.5
c ∗ + c ∗
2h 2l

and

T2h = c∗2h − R(1 − τk )K2∗ − wy2h

∗

T2l = c∗2l − R(1 − τk )K2∗

This way the equilibrium allocations satisfy all of the optimality conditions listed above.
They are also feasible and hence should implement the efficient allocations.
What is missing?
The taxes above guarantee that the individual saves the correct amount “if” he plans to tell
the truth about his ability in the second period. Also, they guarantee that the individual
tells the truth “if” he saves the correct amount.
Question: What if the individual saves more than K2∗ and plan to lie about their ability if
they happen to be the able type?
We first show that this double deviation is feasible and improves individuals welfare. Consider
three different plan by the agent

62
ECN 741: Public Economics Fall 2008

PLAN 1:

• Save K2∗ (≡ K1 + wy1∗ − c∗1 )

∗
• Produce y2h if θ2 = 1 and zero if θ2 = 0

PLAN 2:

• Save K2∗ (≡ K1 + wy1∗ )

• Produce zero in the second period

Note first that, because incentive compatibility in planner’s problem binds, individual is
indifferent between PLAN 1 and PLAN 2. Also,

 
1 .5 .5 1
∗
< R(1 − τk ) ∗ + ∗ < R(1 − τk ) ∗
c1 c2h c2l c2l

PLAN 3:

• Save K2∗ + .

• Consume c∗2l + R(1 − τk ). and produce zero.

Define function g()

g() = log(c∗1 − ) + log(c∗2l + R(1 − τk )) − [log(c∗1 ) + log(c∗2l )]

and note that

1 R(1 − τk )
g 0 (0) = − ∗
+ >0
c1 c∗2l
and therefore

PLAN 3  PLAN 2 ' PLAN 1

Therefore under the proposed tax system individuals prefer PLAN 3 to PLAN 1. Hence this
implementation fails. In other words the efficient allocation cannot be implemented by a
separable tax system in which capital taxes do not depend on individuals income ex-post.

63
ECN 741: Public Economics Fall 2008

3.2.2 Implemeting Efficient Allocations-General Case

We implement efficient allocations using arbitrary nonlinear taxes on labor income and linear
taxes on wealth. The description of envrionment is the following

• Single representative firm that owns the production technology and takes capital rents
rt and wages wt as given.

• Agenst are endowed with K̄1 in period 1.

• Agents trade supply labor and capital to the firm in a sequence of competitive markets.

• Agents face a labor tax schedule ψ : RT+ × Z T → R. ψt is (y t , z t )−measurable. An indi-

vidual with sequence of effectice labor supply {yt }Tt=1 , pays labor taxes ψt ((yt )Tt=1 , z T )
is period t.

• Agents face linear tax on wealth τ : RT+ ×Z T → R. τt is (y t , z t )−measurable. Note that

τt may depend on history of effective labor supply. And individula who holds wealth Wt
and has sequence of effectice labor supply {yt }Tt=1 , pays wealth taxes τt ((yt )Tt=1 , z T )Wt .

Given taz system (ψ, τ ) individual solves

T
yt (θT , z T )
X X X   
T t−1 T T T T
max πz (z ) β πθ (θ |z ) u(ct (θ , z )) − v
c,k,y
t=1
θt
z T ∈Z T θT ∈ΘT

subject to

ct (θT , z T ) + kt+1 (θT , z T ) ≤ 1 − τt (y(θT , z T ), z T ) (1 − δ + rt (z T ))kt (θT , z T )

+w(z T )yt (θT , z T ) − ψt (y(θT , z T ), z T ) ∀(θ, z T )

k1 ≤ K̄1

(ct , kt+1 , yt ) is (θt , z t ) − measurable and nonnegative

Given tax system (ψ, τ ), an eqilibrium is allocations rules (c, k, y) and prices (r, w) such
that given prices and tx system, the allocation rules solves the individual problem, r and
w are such that rt (z T ) = Fkt (Kt (z T ), Yt (z T ), z T ) and wt (z T ) = Fyt (Kt (z T ), Yt (z T ), z T ), and
alloctions are feasible for all t and all z T :

64
ECN 741: Public Economics Fall 2008

Ct (z T ) + Kt+1 (z T ) = (1 − δ)Kt (z T ) + F (Kt (z T ), Yt (z T ), z T )

X
Ct (z T ) = πθ (θT |z T )ct (θT , z T )
θT ∈ΘT
X
Yt (z T ) = πθ (θT |z T )yt (θT , z T )
θT ∈ΘT
X
Kt+1 (z T ) = πθ (θT |z T )kt+1 (θT , z T )
θT ∈ΘT
K1 = K̄1

Note that the tax system described only in terms of sequence of lebor supply (and not skill
shocks). However, households choose consumption as function of past and present skills (the
θ’s). We need to make sure consumption depends on past and present skills only trough past
and present labor supply. In other words, we need to make sure observing past and present
labor supply is enough to determine individual’s consumtion. For that we need to impose
an assumption:
Suppose (c∗ , y ∗ , K ∗ ) is a socially optimal allocation. Define

(y T , z T ) ∈ RT+ × Z T | ∃ θT ∈ ΘT , (ys0 )Ts=t+1 ∈ RT+−t


DOMt =
and (zs0 )Ts=t+1 ∈ Z T −t such that
(y t , (ys0 )Ts=t+1 , z T ) = (y ∗ (θT , z T ), z T )

where z T = (z t , (zs0 )Ts=t+1 ) for some (zs0 )Ts=t+1 ∈ Z T −t .

(y T , z T ) is in DOMt if in socially optimal allocation, there exist a type in ΘT that receives
the effective labor history y t when public history is z t .
Claim: DOMt ⊆ DOMt−1 .
Now we make the following assumption

 T
Assumption 2 There exist a sequence of functions c = cb∗t
b∗ , where cb∗t : DOMt → R+ ,
t=1
cb∗t is (y t , z t )−measurable, and

cb∗t (y ∗ (θT , z T ), z T ) = c∗t (θT , z T ) ∀ (θT , z T )

65
ECN 741: Public Economics Fall 2008

This assumption is satisfied in static case and in i.i.d case. See Kocherlakota (2005) for
further discussion and an example for the case in which this assumption is violated.2
This assumption simply says it is possible to implement consumption allocation as function
of labor supply rather than skill. In this implementation a person who acts like (θT , z T )
(have same labor supply as type (θT , z T )), will also sonsume the same allocation as type
(θT , z T ). Whether this is in fact efficient or not is something we need to show.

Details of the tax system

Given optimal allocation (c∗ , y ∗ , K ∗ ), there exists λ∗t+1 : Z T → R+ (z t+1 −measurable) such
that
 t t+1 −1
   
∗ β 1
λt+1 = 0 ∗ E 0 ∗ θ ,z
u (ct ) u (ct+1 (θt+1 , z̄ t , zt+1 )) 
∗
Define τt+1 : RT+ × Z T → R by

λ∗t+1 (z T )u0 (cb∗t (y ∗ (θT ,z T ))


∗
1 −
βu0 (cd∗ (y ∗ (θ T ,z T ))
∀(y T , z T ) ∈ DOMt+1
τt+1 (y T , z T ) = t+1
1 ∀(y T , z T ) ∈
/ DOMt+1

∗
Note that τt+1 is (y t+1 , z t+1 )−measurable.
Let

M P Kt∗ (z t ) ≡ FK (Kt∗ (z t−1 ), Yt∗ (z t ), z t )

M P L∗t (z t ) ≡ FY (Kt∗ (z t−1 ), Yt∗ (z t ), z t )

Now let (ψ ∗∗ , kb∗ ) : DOMT → RT × RT+ be defined so that

∗ ∗ T
cb∗t (y T , z T ) + kd
∗ T T T T b∗ T T
t+1 (y , z ) = (1 − τt+1 (y , z ))(1 − δ + M P Kt (z ))kt (y , z )

+M P L∗t (z T )yt − ψt∗∗ (y T , z T )

X
∗ ∗ T T T ∗ T
kd
t+1 (y (θ , z ), z ) = Kt (z )
θT

kb1∗ = K1∗

Now we define labor taxes as

2
Page 1601.

66
ECN 741: Public Economics Fall 2008


ψ ∗∗ (y T , z T ) ∀(y T , z T ) ∈ DOM
∗ t t+1
ψt+1 (y T , z T ) =
2y w (z T ) T T
∀(y , z ) ∈ / DOM
t t t+1

Proposition 11 Given prices (r∗ , w∗ ) and tax system (ψ ∗ , τ ∗ ), and given that the typical
agent chooses a budget feasible y 0 , his optimal choices of (c, k) are c0t (θT , z T ) = cb∗t (y 0 (θT , y T ), z T )
and kt0 (θT , z T ) = kbt∗ (y 0 (θT , y T ), z T )

67
ECN 741: Public Economics Fall 2008

References
Atkeson, A. and J. Lucas, Robert E (1992): “On Efficient Distribution with Private
Information,” Review of Economic Studies, 59, 427–53. 3.1.1, 3.1.2

Chamley, C. (1986): “Optimal Taxation of Capital Income in General Equilibrium with

Infinite Lives,” Econometrica, 54, 607–22. 2

Chari, V.V., C. L. and P. Kehoe (1994): “Optimal fiscal policy in a business cycle
model,” Journal of Political Economy, 102, 617–52. 2

Chari, V. and P. J. Kehoe (1998): “Optimal fiscal and monetary policy,” Minneapolis
Fed Staff Report, SR 251. ∗, 2

Corlett, W. and D. Hauge (1953-54): “Complementarity and the excess burden of

taxation,” Review of Economic Studies, 21, 21–30. 1.2.3

Diamond, P. and J. Mirrlees (1971): “Optimal taxation and public production I: pro-
duction efficiency,” American Economic Review, 61, 8–27. 1.4

Ekeland, I. and J. A. Scheinkman (1986): “Transversality Conditions for some infinite

horizon discrete time optimization problems,” Mathematics of operation research, 11. 2

Erosa, A. and M. Gervais (2002): “Optimal Taxation in Life-Cycle Economies,” Journal

of Economic Theory, 105, 338–369. 2.3

Farhi, E. and I. Werning (2005): “Inequality, Social Discounting and Estate Taxation,”
. 3.1.2, 3.1.2

——— (2006): “Progressive Estate Taxation,” . 3.1.2, 3.1.2

——— (2007): “Inequality and Social Discounting,” Journal of Political Economy, 115, 365–
402. 3.1.2, 3.1.2

——— (2008): “Optimal Savings Distortions with Recursive Preferences,” The Journal of
Monetary Economics, 55, 21–42. 3.1.1, 3.1.4

Judd, K. L. (1985): “Redistributive Taxation in a Simple Perfect Foresight Model,” Journal

of Public Economics, 28, 59–83. 2

Kocherlakota, N. R. (2005): “Zero Expected Wealth Taxes: A Mirrlees Approach to

Dynamic Optimal Taxation,” Econometrica, 73, 1587–1621. 3.2.2

68
ECN 741: Public Economics Fall 2008

McGrattan, E. R. and E. C. Prescott (2005): “Taxes, Regulations, and the Value of

U.S. and U.K. Corporations,” Review of Economic Studies, 72, 767–796. 2.2.1

PHELAN, C. (2006): “Opportunity and Social Mobility,” Review of Economic Studies, 73,
487–504. 3.1.2

Sandmo, A. (1987): “A Reinterpretation of Elasticity Formulae in Optimum Tax Theory,”

Economica, 54, 89–96. 1.2.3

Weitzman, M. L. (1973): “Duality Theory of Convex Programming for Infinite Horizon

Economic Models,” Management Science, 19. 2

Werning, I. (2007): “Tax Smoothing with Redistribution,” Quarterly Journal of Eco-

nomics, 122, 925–67. 2.2.3, 2.2.3, 3

69