STOCKHOLM DOCTORAL PROGRAM IN ECONOMICS

Handelshogskolan i Stockholm Stockholms universitet

Paul Klein
Email: [email protected]
URL: http://paulklein.se/dynpub.html
Dynamic Public Finance
Spring 2010
Lecture 1: Ramsey optimal taxation in the growth model
1 The deterministic case
1.1 Preliminaries
Apart from this text, you may want to read Chari and Kehoe (1998).
The problem of optimal taxationhow to raise a given a amount of revenue at least
social costcan be approached in dierent ways. The approach here, following Ramsey
(1927) is to suppose that taxes must be proportional to income and/or consumption. In
particular, lump-sum taxes are ruled out. No theoretical foundations are given for this
particular taxation scheme.
More precisely, the solution concept is the following. The Ramsey optimal allocation is
that allocation that delivers the highest weighted sum of utilities among those allocations
that form part of some competitive equilibrium. Notice that the proportionality of taxes is
built in the concept of competitive equilibrium. Non-proportional taxes mean that people
face dierent after-tax prices, and that is not consistent with competitive equilibrium.
1
A competitive equilibrium consists of an allocation and after-tax prices. Alternatively,
and perhaps more intuitively, we can think of it as having three parts: an allocation,
pre-tax prices and tax rates. In this context it is important to understand that a given
allocation is not associated with a unique set of tax rates. It can easily happen that
the same equilibrium allocation is supported by distinct tax rates. This phenomenon is
known as tax equivalence. To see how this might work, consider an environment where a
representative agent maximizes

t=0

t
u(c
t
,
t
)
subject to
h
t
= 1
t
(1 +
c
t
)c
t
+ k
t+1
= (1
k
t
)r
t
k
t
+ (1
h
t
)w
t
h
t
,
k
0
given and some suitable No Ponzi Scheme constraint.
The agents rst order conditions are
(1
h
t
)w
t
u
c,t
= (1 +
c
t
)u
,t
and
(1 +
c
t+1
)u
c,t
= (1 +
c
t
)u
c,t+1
(1
k
t+1
)r
t+1
.
Now imagine an equilibrium where
c
t
= 0,
h
t
=
h
and
k
t
=
k
. Suppose now that
capital taxes are deemed unconstitutional so that
k
t
= 0. Can we now redesign the other
taxes so that the previous equilibrium allocation is still an optimal choice?
Heres how to do it. In the rst equation, what matters is the ratio (1
h
t
)/(1 +
c
t
).
Lets make sure that this is unchanged, so that
1
h
t
1 +
c
t
= 1
h
and, similarly,
1 +
c
t+1
1 +
c
t
=
1
1
k
.
2
Clearly this leads to no contradiction. Indeed, we even have an extra degree of freedom.
Set
c
0
arbitrarily; for instance, set it to zero. To satisfy the second equation, we have to
set
1 +
c
t+1
=
1 +
c
t
1
k
or, when
c
0
,
1 +
c
t
= (1
k
)
t
.
If
k
> 0 this implies an ever increasing consumption tax. Meanwhile,
1
h
t
= (1 +
c
t
)(1
h
).
You may wonder if this tilde tax scheme is truly equivalent to the original one in the sense
that it balances the governments intertemporal budget constraint. Well, with an arbitrary
choice of
c
0
it doesnt necessarily. But if we are free to manipulate that toousing up
that one degree of freedomwe can use it to balance the governments intertemporal
budget.
1.2 The primal approach
I hope that the previous section convinced you that it is better to try to eliminate taxes
and go for the Ramsey optimal allocation directly, without taking a stand on precisely
how that allocation is enforced as a competitive equilibrium. The question is how to do
that.
In a way, the dual approach is more straightforward. Let allocations be a function of tax
rates (via individual optimality, market clearing and government budget balance). Then
maximize utility with respect to the tax rates. The lack of uniqueness of the solution
makes this impractical, but it is conceptually straightforward.
Our path will be more roundabout but nevertheless better. What we will do is to char-
acterize the set of allocations that can be supported as a competitive equilibriumfor
3
which there exist prices that, together with the given allocation, constitutes a competitive
equilibrium.
Every allocation is not a competitive equilibrium allocation in this sense, even if it is
resource feasible. To see this, consider what must be true of every competitive equilibrium.
Denote by p
t
the after-tax price of good t in terms of good 0 and by w
t
the after-tax relative
price of leisure relative to consumption. We normalize p
0
= 1.

t=0
p
t
(c
t
w
t
h
t
) = a
0
where a
0
is the after-tax value of initial assets. We will see later that the government
would like to tax these initial assets, but that amounts to a lump-sum tax which is (at least
pragmatically) inconsistent with the idea that taxes have to be proportional to income
or consumption (as opposed to assets). For this reason, it is tempting to regard a
0
as a
parameter; otherwise, the optimal taxation problem might easily become trivial. Below
we will see that it would be awkward to take this idea too literally and a
0
will in fact be
subject to some manipulation.
Incidentally, a nice thing about the present value formulation is that we dont have to
bother with an explicit treatment of government debt. All we need to know at this stage
is that its allowed. If we were really interested in it, we could back it out from the
allocation and the rst order conditions. (A perfect homework exercise!)
With this after-tax budget constraint, the rst order conditions are simplied quite a bit.
They are
w
t
u
c,t
= u
,t
and
p
t+1
p
t
=
u
c,t+1
u
c,t
.
The second equation means that we can write
p
t
=
t
u
c,t
u
c,0
.
4
The rst (obviously) implies
w
t
=
u
,t
u
c,t
.
Thus the consumers budget constraint becomes

t=0

t
(u
c,t
c
t
u
,t
h
t
) = u
c,0
a
0
. (1)
Evidently this equation must be satised if an allocation is to be a part of any competitive
equilibrium. We call it the implementability condition. The classic statement of this
condition is in Lucas and Stokey (1983).
It turns out that something close to the converse is true as well: if an allocation satises (1)
and the resource constraint, then the allocation is part of some competitive equilibrium.
To show this, all we need to do is to come up with after-tax prices that are consistent
with the given allocation being an optimal choice and the consumers intertemporal budget
constraint is satised.
What about the governments budget constraint? It is implied by the resource constraint
and the consumers budget constraint, taken together. There is no need to impose all
three. Now, then, is the time to introduce the resource constraint...s. There is one for
every period, and this is signicant.
c
t
+ g
t
+ k
t+1
= f(k
t
, h
t
)
where the sequence (g
t
) is exogenously given.
Now suppose we have an allocation, i.e. a sequence (c
t
) and a sequence (h
t
). What about
k
t
? We can back that out from the resource constraint, given k
0
. What is k
0
? Apparently
we have to take not only a
0
as given but also k
0
. The most straightforward way of making
these assumptions is to take initial government debt b
0
, initial capital k
0
and the initial
capital income tax
k
0
as given. Initial asset holdings then follow via
a
0
= (1
k
0
)r
0
(b
0
+ k
0
).
5
Notice that the value of this expression can in fact be manipulated by the government
through h
0
which aects r
0
.
We now interrupt our discussion for a conceptual question. What did I mean when I said
that an allocation must satisfy the resource constraint? Since we back out (k
t
) from the
resource constraint, it would appear that every allocation satises the resource constraint
for some appropriately chosen sequence (k
t
). Oh no it doesnt! This is because we must
have k
t
0 for all t. If an allocation (c
t
), (h
t
) is implies k
t+1
< 0 for some t then we say
that it is not resource feasible.
Now lets construct after-tax prices so that our resource feasible allocation is part of a
competitive equilibrium. This is almost ridiculously easy.
w
t
=
u
,t
u
c,t
and
p
t
=
t
u
c,t
u
c,0
.
1.3 The Chamley-Judd result
Independently of one another, Chamley (1986) and Judd (1985) showed that intertemporal
choices should not be distorted in the long run. That result may be described by saying
that capital should not be taxed. But given tax equivalence, zero capital taxes involve
no loss of generality so to say that capital should not be taxed is a vacuous statement.
Meanwhile, the Chamley-Judd result is not vacuous. What it says is that, in the limit as
t we should have
u
c,t
= u
c,t+1
f
k,t+1
.
To see this, consider the Ramsey optimal taxation problem. It is to maximize

t=0

t
u(c
t
,
t
)
6
subject to
h
t
= 1
t
,
c
t
+ k
t+1
= f(k
t
, h
t
),
k
t+1
0,

t=0

t
[u
c,t
c
t
u
,t
h
t
] = u
c,0
(1
k
0
)r
0
(b
0
+ k
0
)
and k
0
and a
0
given.
Notice that although initial assets a
0
and k
0
are xed, their real value in terms of
current goods can be reduced by reducing r
0
= f
k
(k
0
, h
0
), i.e. by encouraging an initial
day of rest. Moreover, their value in terms of future goods can be reduced by reducing
u
c,0
, i.e. encouraging an initial binge. More about that later.
Now denote the Lagrange multiplier associated with the implementability constraint
by and dene
W(c, h; ) = u(c, 1 h) + [u
c
+ u

h].
Now form the Lagrangian:
L =

t=0

t
W(c
t
, h
t
) +

t=0

t
[f(k
t
, h
t
) c
t
k
t+1
] + u
c,0
(1
k
0
)r
0
(b
0
+ k
0
)
The rst order conditions with respect to c
t
are, for all t = 0,
W
c,t
=
t
and with respect to k
t+1
they are

t
= f
k,t+1

t+1
.
Combining these familiar-looking results, we have the following equally familiar-looking
result.
W
c,t
= f
k,t+1
W
c,t+1
.
7
Now suppose consumption, leisure and the capital stock all converge to constants in the
long run. Then W
c,t
= W
c,t+1
and
1 = f
k
.
Evidently this is exactly the same optimality condition as in the Pareto optimal allocation.
Under some conditions, intertemporal choices are not distorted as of period 1. The is true
(why?) if
W
c,t
u
c,t
=
W
c,t+1
u
c,t+1
.
Under what conditions might this be true? Suppose
u(c
t
,
t
) =
c
1
t
1
+ v(
t
). (2)
Then
W
c,t
= [1 + (1 )]u
c,t
and the condition is satised.
1.4 Backing out taxes and debt from the allocation
Given an allocation, it is often interesting to back out the policy that implements it. This
cannot be done uniquely. To make the policy unique, it is necessary to impose restrictions.
The meta-theorem is that you need precisely as many policy instruments as there are
choice variables.
In the growth model there are three choice variables: consumption, leisure and investment.
So let there be three instruments: a labor tax, a capital tax and government debt. (It
matters that we allow for government debt. On the other hand, as we have seen, it doesnt
matter whether we allow for capital taxes or consumption taxes.)
Consider the consumers rst order condition for labor supply, evaluated in a competitive
8
equilibrium where pre-tax wages are equal to the marginal products of labor.
u
,t
= (1
h
t
)f
h,t
u
c,t
.
This means that

h
t
= 1
u
,t
u
c,t
f
h,t
.
Meanwhile,
u
c,t
= (1
k
t+1
)f
k,t+1
u
c,t+1
so that, for t > 0,

k
t
= 1
u
c,t1
f
k,t
u
c,t
.
Finally, we can back out debt via
b
t+1
=
1
(1
k
t
)f
k,t
b
t
+ g
t

h
t
f
h,t
h
t

k
t
f
k,t
k
t
.
1.5 Computation
What is the correct value of ? Set = 0 and solve for the optimal allocation. Plug it into
Equation (1) and see if it is satised. (If g
t
0 it will be, otherwise there is something
wrong.) If not, increase a little bit and solve again, using your previous solution as an
excellent initial guess. Keep going until Equation (1) is satised.
How to solve the model for a given ? Heres the best way to do that when the model is
deterministic. What we do is treat all the equilibrium conditions as a system of nonlinear
equations and solve this system. Of course, there are innitely many such equations. To
deal with this problem, we assume that the steady state is reached after T periods, which
is an excellent approximation if T is big enough.
The approach starts by computing the steady state. This is just a matter of solving a
rather low-dimensional system of non-linear equations.
9
After that, solve for the transition. When computing it, it is useful to distinguish between
static and dynamic equilibrium conditions. To dene what I mean by that, denote
the state by x
t
and the other variables by d
t
. The static equilibrium conditions dont
involve d
t+1
, and lets say that there are n
d
of them. The dynamic conditions do involve
d
t+1
and there are n
x
of them.
More precisely, let the equilibrium conditions be given by

f(x
t
, x
t+1
, d
t
, d
t+1
) = 0
g(x
t
, x
t+1
, d
t
) = 0
With this notation in place, the equilibrium conditions are

g(x
0
, x
1
, d
0
) = 0
f(x
0
, x
1
, d
0
, d
1
) = 0
g(x
1
, x
2
, d
1
) = 0
f(x
1
, x
2
, d
1
, d
2
) = 0
.
.
.
f(x
T1
, x
T
, d
T1
, d
T
) = 0
g(x
T
, x
T+1
, d
T
) = 0
where x
0
is xed by an initial condition and x
T+1
is xed by the steady state. What
remains to solve for is x
1
, x
2
, . . . , x
T
and d
0
, d
1
, . . . , d
T
. Notice that there are just as many
equations as there are unknowns!
1.6 Balancing the budget
What if government debt is not allowed. This is a binding constraint. How do we impose
it? After all, debt does not feature at all in the primal approach, so how on earth do we
10
set it to zero? By translating into terms involving allocations only.
Apparently the budget is balanced if
g
t
=
h
t
f
h,t
h
t
+
k
t
f
k,t
k
t
Translating, we get
g
t
= [1
u
,t
u
c,t
f
h,t
]f
h,t
h
t
+ [1
u
c,t1
f
k,t
u
c,t
]f
k,t
k
t
Assuming constant returns to scale so that y
t
= f(k
t
, h
t
) = f
k,t
k
t
+ f
h,t
h
t
, we have
u
c,t
(y
t
g
t
) = u
,t
h
t
+
1
u
c,t1
k
t
.
Equivalently,
u
c,t
k
t
= [u
c,t+1
(c
t+1
+ k
t+1
) u
t
h
t
].
This can now be imposed as a constraint. Its a non-standard constraint from the point of
view of dynamic optimization theory. Specically, it cannot be written as x
t+1
= g(x
t
, d
t
).
This raises some technical issues if you are interested in nding a feedback solution, which
is the only feasible option if there is uncertainty. These technical issues are addressed, at
a rather high level of generality, in Marcet and Marimon (2009).
2 Optimal taxation under uncertainty
2.1 With state-contingent debt
Consider an environment without capital where a representative consumer maximizes
E

t=0

t
u(c
t
,
t
)

11
and where the resource constraint is
c
t
+ g
t
= h
t
= 1
t
where {g
t
} is a stochastic process living on the probability space (, F, P). Let {F
t
} be
the ltration generated by {g
t
}. Markets are complete so that all contingent claims are
associated with a well-dened price. Moreover, there are no limits on the quantities of
contingent claims held in any particular period. Let {Q
t
: F R
+
} be a sequence of price
measures with the following interpretation. The contingent claim that delivers 1 unit of
consumption in period t if A has price Q
t
(A). We normalize so that Q
0
() = 1. Let
W
t
be the (after-tax) leisure price measure.
Meanwhile, dene the stochastic processes p
t
and w
t
via
dQ
t
= p
t
dP on F
t
and
dW
t
= w
t
dP on F
t
.
With these denitions, the consumers budget constraint (conceived as an equality) can
be written as
E

t=0
[p
t
c
t
w
t
h
t
]

= b
0
.
By the consumers optimality conditions,
p
t
=
t
u
c,t
u
c,0
and
w
t
=
t
u
h,t
u
c,0
.
Hence the implementability condition becomes
E

t=0

t
[u
c,t
c
t
u
,t
h
t
]

= u
c,0
b
0
.
12
The equivalent sequence-of-markets representation of this condition is to assert the exis-
tence of an {F
t
}-adapted stochastic process {b
t
} with initial value b
0
and which satises
E[u
c,t+1
b
t+1
|F
t
] + u
c,t
c
t
u
,t
h
t
= u
c,t
b
t
for all t = 0, 1, 2, . . . and
lim
t

t
E[u
c,t
b
t
] = 0.
2.2 Without debt
The balanced budget constraint in this case becomes:
u
c,t
k
t
= E[ u
c,t+1
(c
t+1
+ k
t+1
) u
,t+1
h
t+1
| F
t
] .
Can you derive it?
2.3 Without state-contingent debt
Aiyagari et al. (2002), in a paper based on Marcet et al. (1996), consider a world where
debt is not state-contingent, i.e. a government bond must be a riskless asset whose return
is determined one period in advance. This topic is also covered in Ljungqvist and Sargent
(2000).
It is natural to impose these constraints on the sequence-of-markets representation rather
than the present-value representation. We had
E[u
c,t+1
b
t+1
|F
t
] + u
c,t
c
t
u
,t
h
t
= u
c,t
b
t
and we now impose
E[b
t+1
|F
t
] = b
t+1
.
13
We get
E[u
c,t+1
|F
t
]b
t+1
+ u
c,t
c
t
u
,t
h
t
= u
c,t
b
t
.
In order for this predictability of debt to have any bite, we also impose
b b
t+1
b
for each t = 0, 1, 2, . . . Incidentally, this guarantees
lim
t

t
E[u
c,t
b
t
] = 0.
A particularly simple case is
u(c, h) = c
1
2
h
2
t
where is some increasing convex function. With these preferences we will be able to
show that labour taxes should follow a random walk. Since taxes dont feature explicitly
in our analysis we will instead show, equivalently, that the marginal utility of leisure is a
martingaleat least as long as the debt constraints dont bind.
With these preferences, the governments budget constraint is
b
t+1
+ c
t
h
2
t
= b
t
.
Dene the Hamiltonian via
H =
t
[c
t

1
2
h
2
t
] +
1

t+1
[b
t
c
t
+ h
2
t
].
Subsituting in the resource constraint, we get
H =
t
[h
t
g
t

1
2
h
2
t
] +
1

t+1
[b
t
h
t
+ g
t
+ h
2
t
].
The optimality conditions are

t
[1 h
t
] +
1
E
t
[
t+1
](1 + 2h
t
)
and
E
t
[
t+1
] =
t
.
14
Now dene

t
=
t

t
.
We get
1 h
t
+E
t
[
t+1
](1 + 2h
t
) = 0
and
E
t
[
t+1
] =
t
which means that
t
is a martingale. Meanwhile, we have
1 h
t
+
t
(1 2h
t
) = 0.
Solving for h
t
, we get
h
t
=
1 +
t
1 + 2
t
.
What about the labour income tax rate? Apparently

h
t
= 1 h
t
and so

h
t
=

t
1 + 2
t
so that although the tax rate is not a martingale, it is a function of a martingale and
therefore quite persistent, certainly more persistent than the underlying shock.
15
References
Aiyagari, S. R., A. Marcet, T. Sargent, and J. Seppala (2002). Optimal taxation without
state-contingent debt. Journal of Political Economy 110, 12201254.
Chamley, C. (1986). Optimal taxation of capital income in general equilibrium with
innite lives. Econometrica 54, 607622.
Chari, V. V. and P. J. Kehoe (1998). Optimal scal and monetary policy. Federal Reserve
Bank of Minneapolis Research Department Sta Report 251.
Judd, K. L. (1985). Redistributive taxation in a simple perfect foresight model. Journal
of Public Economics 28, 59:83.
Ljungqvist, L. and T. Sargent (2000). Recursive Macroeconomic Theory. MIT Press.
Lucas, R. E. and N. L. Stokey (1983). Optimal scal and monetary policy in an economy
without capital. Journal of Monetary Economics 12(1), 5593.
Marcet, A. and R. Marimon (2009). Recursive contracts. Manuscript.
Marcet, A., T. J. Sargent, and J. Seppal a (1996). Optimal taxation without state-
contingent debt. Manuscript.
Ramsey, F. P. (1927, March). A contribution to the theory of taxation. Economic Jour-
nal 37(145), 4761.
16