Solving The Intertemporal Consumption/saving Problem in Discrete and Continuous Time

Chapter 9
Solving the intertemporal

consumption/saving problem in
discrete and continuous time
In the next two chapters we shall discuss the continuous-time version of the
basic representative agent model, the Ramsey model, and some of its ap-
plications. As a preparation for this, the present chapter gives an account
of the transition from discrete time to continuous time analysis and of the
application of optimal control theory for solving the households consump-
tion/saving problem in continuous time.
There are many elds of study where a setup in continuous time is prefer-
able to one in discrete time. One reason is that continuous time opens up
for application of the mathematical apparatus of dierential equations; this
apparatus is more powerful than the corresponding apparatus of dierence
equations. Another reason is that optimal control theory is more developed
and potent in its continuous time version than in its discrete time version,
considered in Chapter 8. In addition, many formulas in continuous time are
simpler than the corresponding ones in discrete time (cf. the growth formulas
in Appendix A).
As a vehicle for comparing continuous time analysis with discrete time
analysis we consider a standard household consumption/saving problem.
How does the household assess the choice between consumption today and
consumption in the future? In contrast to the preceding chapters we allow
for an arbitrary number of periods within the time horizon of the household.
The period length may thus be much shorter than in the previous models.
This opens up for capturing additional aspects of economic behavior and for
carrying out the transition to continuous time in a smooth way.
First, we shall specify the market environment in which the optimizing
327
328
CHAPTER 9. SOLVING THE INTERTEMPORAL CONSUMPTION/-
SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME
household operates.
9.1 Market conditions
In the Diamond OLG model no loan market was active and wealth eects on
consumption through changes in the interest rate were absent. It is dierent
in a setup where agents live for many periods and realistically have a hump-
shaped income prole through life. This motivates a look at the nancial
market and more rened notions related to intertemporal choice.
We maintain the assumption of perfect competition in all markets, i.e.,
households take all prices as given from the markets. Ignoring uncertainty,
the various assets (real capital, stocks, loans etc.) in which households invest
give the same rate of return in equilibrium. To begin with we consider time
as discrete.
A perfect loan market Consider a given household. Suppose it can at
any date take a loan or provide loans to others at the going interest rate, i
t
,
measured in money terms. That is, monitoring, administration, and other
transaction costs are absent so that (a) the household faces the same interest
rate whether borrowing or lending; (b) the household can not inuence this
rate; and (c) there are no borrowing restrictions other than the requirement
on the part of the borrower to comply with her nancial commitments. The
borrower can somehow be forced to repay the debt with interest and so
the lender faces no default risk. A loan market satisfying these idealized
conditions is called a perfect loan market. The implications of such a market
are:
1. various payment streams can be subject to comparison; if they have
the same present value (PV for short), they are equivalent;
2. any payment stream can be converted into another one with the same
present value;
3. payment streams can be compared with the value of stocks.
Consider a payment stream {r
t
}
T1
t=0
over 1 periods, where r
t
is the pay-
ment in currency at the end of period t. As in the previous chapters, period
t runs from time t to time t +1 for t = 0. 1. .... 1 1; and i
t
is dened as the
interest rate on a loan from time t to time t+1. Then the present value, 1\
0
,
c Groth, Lecture notes in macroeconomics, (mimeo) 2013.
9.1. Market conditions 329
as seen from the beginning of period 0, of the payment stream is dened as
1
1\
0
=
r
0
1 + i
0
+
r
1
(1 + i
0
)(1 + i
1
)
+ +
r
T1
(1 + i
0
)(1 + i
1
) (1 + i
T1
)
. (9.1)
If Ms. Jones is entitled to the income stream {r
t
}
T1
t=0
and at time 0
wishes to buy a durable consumption good of value 1\
0
, she can borrow this
amount and use the income stream{r
t
}
T1
t=0
to repay the debt over the periods
t = 0. 1. 2. .... 1 1. In general, when Jones wishes to have a time prole
on the payment stream dierent from the income stream, she can attain this
through appropriate transactions in the loan market, leaving her with any
stream of payments of the same present value as the income stream.
The good which is traded in the loan market will here be referred to
as a bond. The borrower issues bonds and the lender buys them. In this
chapter all bonds are assumed to be short-term, i.e., one-period bonds. For
every unit of account borrowed in the form of a one-period loan at the end of
period t 1, the borrower pays back with certainty (1 + short-term interest
rate) units of account at the end of period t. If a borrower wishes to maintain
debt through several periods, new bonds are issued and the obtained loans
are spent rolling over the older loans at the going market interest rate. For
the lender, who lends in several periods, this is equivalent to a variable-rate
demand deposit in a bank.
2
Real versus nominal rate of return As in the preceding chapters our
analysis will be in real terms, also called ination-corrected terms. In princi-
ple the unit of account is a xed bundle of consumption goods. In the simple
macroeconomic models to be studied in this and subsequent chapters, such
a bundle is reduced to one consumption good because the models assume
there is only one consumption good in the economy. Moreover, there will
only be one produced good, the output good, which can be used for both
consumption and capital investment. Then, whether we say our unit of ac-
count is the consumption good or the output good does not matter. To x
our language, we will say the latter.
The real (net) rate of return on an investment is the rate of return in
units of the output good. More precisely, the real rate of return in period t.
:
t
. is the (proportionate) rate at which the real value of an investment, made
at the end of period t 1. has grown after one period.
1
We use present value as synonymous with present discounted value. As usual
our timing convention is such that 1\
0
denotes the time-0 value of the payment stream,
including the discounted value of the payment (or dividend) indexed by 0.
2
Unless otherwise specied, we use terms like loan market, credit market, and
bond market interchangeably.
330
The link between this rate of return and the more commonplace concept
of a nominal rate of return is the following. Imagine that at the end of period
t 1 you make a deposit of value \
t
euro on an account in a bank. The real
value of the deposit when you invest is then \
t
,1
t1
. where 1
t1
is the price
in euro of the output good at the end of period t 1. If the nominal short-
term interest rate is i
t
. the deposit is worth \
t+1
= \
t
(1 +i
t
) euro at the end
of period t. By denition of :
t
. the factor by which the deposit in real terms
has expanded is
1 + :
t
=
\
t+1
,1
t
\
t
,1
t1
=
\
t+1
,\
t
1
t
,1
t1
=
1 + i
t
1 + :
t
. (9.2)
where :
t
(1
t
1
t1
),1
t1
is the ination rate in period t. So the real (net)
rate of return on the investment is :
t
= (i
t
:
t
),(1 +:
t
) i
t
:
t
for i
t
and
:
t
small. The number 1+:
t
is called the real interest factor and measures
the rate at which current units of output can be traded for units of output
one period later.
In the remainder of this chapter we will think in terms of real values and
completely ignore monetary aspects of the economy.
9.2 Maximizing discounted utility in discrete
time
We assume that the consumption/saving problem faced by the household
involves only one consumption good. So the composition of consumption in
each period is not part of the problem. What remains is the question how to
distribute consumption over time.
The intertemporal utility function
A plan for consumption in the periods 0. 1. .... 1 1 is denoted {c
t
}
T1
t=0
,
where c
t
is the consumption in period t. We say the plan has time horizon
1. We assume the preferences of the household can be represented by a time-
separable intertemporal utility function with a constant utility discount rate
and no utility from leisure. The latter assumption implies that the labor
supply of the household in each period is inelastic. The time-separability
itself just means that the intertemporal utility function is additive, i.e.,
l(c
0
. c
1
.. . . . c
T1
) = n
(0)
(c
0
) + n
(1)
(c
1
)+ . . . +n
(T1)
(c
T1
). where n
(t)
(c
t
) is
the utility contribution from period-t consumption, t = 0. 1.. . . . 1 1. But in
addition we assume there is a constant utility discount rate j 1. implying
9.2. Maximizing discounted utility in discrete time 331
that n
(t)
(c
t
) = n(c
t
)(1 +j)
t
. where n(c) is a time-independent period utility
function. Together these two assumptions amount to
l(c
0
. c
1
. . c
T1
) = n(c
0
)+
n(c
1
)
1 + j
+ . . . +
n(c
T1
)
(1 + j)
T1
=
T1
X
t=0
n(c
t
)
(1 + j)
t
. (9.3)
The period utility function is assumed to satisfy n
0
(c) 0 and n
00
(c) < 0.
Box 9.1. Admissible transformations of the period utility function
When preferences, as assumed here, can be represented by discounted utility, the
concept of utility appears at two levels. The function l() in (9.3) is dened on
the set of alternative feasible consumption paths and corresponds to an ordinary
utility function in general microeconomic theory. That is, l() will express the
same ranking between alternative consumption paths as any increasing transfor-
mation of l(). The period utility function, n(), dened on the consumption in
a single period, is a less general concept, requiring that reference to utility
units is legitimate. That is, the size of the dierence in terms of period utility
between two outcomes has signicance for choices. Indeed, the essence of the
discounted utility hypothesis is that we have, for example,
n(c
0
) n(c
0
0
) 0.95
n(c
0
1
) n(c
1
)
(c
0
. c
1
) (c
0
0
. c
0
1
).
meaning that the household, having a utility discount factor 1,(1 + j) = 0.95,
strictly prefers consuming (c
0
. c
1
) to (c
0
0
. c
0
1
) in the rst two periods, if and only
if the utility dierences satisfy the indicated inequality. (The notation r
means that r is strictly preferred to .)
Only a linear positive transformation of the utility function n(). that is,
(c) = cn(c) + /, where c 0, leaves the ranking of all possible alternative
consumption paths, {c
t
}
T1
t=0
, unchanged. This is because a linear positive
transformation does not aect the ratios of marginal period utilities (the mar-
ginal rates of substitution across time).
To avoid corner solutions we impose the No Fast Assumption lim
c0
n
0
(c) =
. As (9.3) indicates, the number 1 +j tells how many extra units of utility
in the next period the household insists on to compensate for a decrease of
one unit of utility in the current period. So, a j 0 will reect that if the
chosen level of consumption is the same in two periods, then the individual
always appreciates a marginal unit of consumption higher if it arrives in the
earlier period. This explains why j is named the rate of time preference or
even more to the point the rate of impatience. The utility discount factor,
332
1,(1+j)
t
, indicates how many units of utility the household is at most willing
to give up in period 0 to get one additional unit of utility in period t.
3
It is generally believed that human beings are impatient and that j should
therefore be assumed positive; indeed, it seems intuitively reasonable that the
distant future does not matter much for current private decisions.
4
There
is, however, a growing body of evidence suggesting that the utility discount
rate is generally not constant, but declining with the time distance from
the current period to the future periods within the horizon. Since this last
point complicates the models considerably, macroeconomics often, as a rst
approach, ignores it and assumes a constant j to keep things simple. Here
we follow this practice. Except where needed, we shall not, however, impose
any other constraint on j than the denitional requirement in discrete time
that j 1.
As explained in Box 9.1, only linear positive transformations of the period
utility function are admissible.
The saving problem in discrete time
Suppose the household considered has income from two sources: work and
nancial wealth. Let c
t
denote the real value of nancial wealth held by the
household at the beginning of period t (c for assets). We treat c
t
as pre-
determined at time t and in this respect similar to a variable-interest deposit
in a bank. The initial nancial wealth, c
0
. is thus given, independently of
whatever might happen to expected future interest rates. And c
0
can be
positive as well as negative (in the latter case the household is initially in
debt).
The labor income of the household in period t is denoted n
t
0 and may
follow a typical life-cycle pattern with labor income rst rising, then more
or less stationary, and nally vanishing due to retirement. Thus, in contrast
to previous chapters where n
t
denoted the real wage per unit of labor, here
a broader interpretation of n
t
is allowed. Whatever the time prole of the
amount of labor delivered by the household through life, in this chapter,
where the focus is on individual saving, we regard this time prole, as well as
the hourly wage as exogenous. The present interpretation of n
t
will coincide
3
Multiplying through in (9.3) by (1+j)
1
would make the objective function appear in
a way similar to (9.1) in the sense that also the rst term in the sum becomes discounted.
At the same time the ranking of all possible alternative consumption paths would remain
unaected. For ease of notation, however, we use the form (9.3) which is more standard.
4
If uncertainty were included in the model, (1 + j)
1
might be seen as reecting the
probability of surviving to the next period and in this perspective j 0 seems a plausible
assumption.
with the one in the other chapters if we imagine that the household in each
period delivers one unit of labor.
Since uncertainty is by assumption ruled out, the problem is to choose a
plan (c
0
. c
1
.. . . . c
T1
) so as to maximize
l =
T1
X
t=0
n(c
t
)(1 + j)
t
s.t. (9.4)
c
t
0. (9.5)
c
t+1
= (1 + :
t
)c
t
+ n
t
c
t
. c
0
given. (9.6)
c
T
0. (9.7)
where :
t
is the interest rate. The control region (9.5) reects the denitional
non-negativity of consumption. The dynamic equation (9.6) is an accounting
relation telling how nancial wealth moves over time. Indeed, income in
period t is :
t
c
t
+ n
t
and saving is then :
t
c
t
+n
t
c
t
. Since saving is by
denition the same as the increase in nancial wealth, c
t+1
c
t
. we obtain
(9.6). Finally, the terminal condition (9.7) is a solvency requirement that
no nancial debt be left over at the terminal date, 1. We shall refer to this
decision problem as the standard discounted utility maximization problem
without uncertainty.
Solving the problem
To solve the problem, let us use the substitution method.
5
From (9.6) we
have c
t
= (1 + :
t
)c
t
+ n
t
c
t+1
. for t = 0. 1.. . . . 1 1. Substituting this
into (9.4), we obtain a function of c
1
. c
2
.. . . . c
T
. Since n
0
0. saturation is
impossible and so an optimal solution cannot have c
T
0. Hence we can put
c
T
= 0 and the problem is reduced to an essentially unconstrained problem of
maximizing a function

l w.r.t. c
1
. c
2
.. . . . c
T1
. Thereby we indirectly choose
c
0
. c
1
.. . . . c
T2
. Given c
T1
. consumption in the last period is trivially given
as
c
T1
= (1 + :
T1
)c
T1
+ n
T1
.
ensuring c
T
= 0.
To obtain rst-order conditions we put the partial derivatives of

l w.r.t.
c
t+1
, t = 0. 1.. . . . 1 2. equal to 0:
J

l
Jc
t+1
= (1 + j)
t
n
0
(c
t
) (1) + (1 + j)
1
n
0
(c
t+1
)(1 + :
t+1
)
= 0.
5
Alternative methods include the Maximum Principle as described in the previous
chapter or Dynamic Programming as described in the appendix to Chapter 29.
334
Reordering gives the Euler equations describing the trade-o between con-
sumption in two succeeding periods,
n
0
(c
t
) = (1 + j)
1
n
0
(c
t+1
)(1 + :
t+1
). t = 0. 1. 2. .... 1 2. (9.8)
Interpretation The interpretation of (30.14) is as follows. Let the con-
sumption path (c
0
. c
1
.. . . . c
T1
) be our reference path. Imagine an alter-
native path which coincides with the reference path except for the periods
t and t + 1. If it is possible to obtain a higher total discounted utility than
in the reference path by varying c
t
and c
t+1
within the constraints (9.5),
(9.6), and (9.7), at the same time as consumption in the other periods is
kept unchanged, then the reference path cannot be optimal. That is, local
optimality is a necessary condition for global optimality. So the optimal
plan must be such that the current utility loss by decreasing consumption c
t
by one unit equals the discounted expected utility gain next period by having
1 + :
t+1
extra units available for consumption, namely the gross return on
saving one more unit in the current period.
A more concrete interpretation, avoiding the notion of utility units, is
obtained by rewriting (30.14) as
n
0
(c
t
)
(1 + j)
1
n
0
(c
t+1
)
= 1 + :
t+1
. (9.9)
The left-hand side indicates the marginal rate of substitution, MRS, of period-
(t +1) consumption for period-t consumption, namely the increase in period-
(t + 1) consumption needed to compensate for a one-unit marginal decrease
in period-t consumption:
`1o
t+1,t
=
dc
t+1
dc
t
|
l=
l
=
n
0
(c
t
)
(1 + j)
1
n
0
(c
t+1
)
. (9.10)
And the right-hand side of (9.9) indicates the marginal rate of transformation,
MRT, which is the rate at which the loan market allows the household to
shift consumption from period t to period t + 1. In an optimal plan MRS
must equal MRT.
The formula (9.10) for MRS indicates why the assumption of a constant
utility discount rate is convenient (but also restrictive). The marginal rate of
substitution between consumption this period and consumption next period
is independent of the level of consumption as long as this level is the same
in the two periods.
Moreover, the formula for MRS between consumption this period and
consumption two periods ahead is
`1o
t+2,t
=
dc
t+2
dc
t
|
l=
l
=
n
0
(c
t
)
(1 + j)
2
n
0
(c
t+2
)
.
This displays one of the reasons that the time-separability of the intertem-
poral utility function is a strong assumption. It implies that the trade-o
between consumption this period and consumption two periods ahead is in-
dependent of consumption in the interim.
Deriving a consumption function The rst-order conditions (30.14) tell
us about the relative consumption levels over time, not the absolute level.
The latter is determined by the condition that initial consumption, c
0
. must
be highest possible, given that the rst-order conditions and the constraints
(9.6) and (9.7) must be satised.
To nd an explicit solution we have to specify the period utility function.
As an example we choose the CRRA function n(c) = c
10
,(1 o). where
o 0.
6
Moreover we simplify by assuming :
t
= :. a constant 1. Then
the Euler equations take the form (c
t+1
,c
t
)
0
= (1 + :)(1 + j)
1
so that
c
t+1
c
t
=

1 + :
1 + j
10
. (9.11)
and thereby c
t
=
t
c
0
. t = 0. 1.. . . . 1 1. Substituting into the accounting
equation (9.6), we thus have c
t+1
= (1 + :)c
t
+ n
t

t
c
0
. By backward
substitution we nd the solution of this dierence equation to be
c
t
= (1 + :)
t
"
c
0
+
t1
X
i=0
(1 + :)
(i+1)
(n
i
i
c
0
)
#
.
Optimality requires that the left-hand side of this equation vanishes for
t = 1. So we can solve for c
0
:
c
0
=
1 + :
P
T1
i=0

1+v
i
"
c
0
+
T1
X
i=0
(1 + :)
(i+1)
n
i
#
=
1 + :
P
T1
i=0

1+v
i
(c
0
+ /
0
).
(9.12)
where we have inserted the human wealth of the household (present value of
expected lifetime labor income) as seen from time zero:
/
0
=
T1
X
i=0
(1 + :)
(i+1)
n
i
. (9.13)
6
In later sections of this chapter we will let the time horizon of the decision maker go to
innity. To ease convergence of an innite sum of discounted utilities, it is an advantage
not to have to bother with additive constants in the period utilities and therefore we write
the CRRA function as c
1
,(1 0) instead of the form, (c
1
1),(1 0), introduced in
Chapter 3. As implied by Box 9.1, the two forms represent the same preferences.
336
Thus (9.12) says that initial consumption is proportional to initial total
wealth, the sum of nancial wealth and human wealth at time 0. To allow
positive consumption we need c
0
+ /
0
0. This can be seen as a solvency
condition which we assume satised.
As (9.12) indicates, the propensity to consume out of total wealth depends
on:
T1
X
i=0

1 + :
i
=
(
1
(

1+r
)
T
1

1+r
when 6= 1 + :.
1 when = 1 + :.
(9.14)
where the result for 6= 1+: follows from the formula for the sum of a nite
geometric series. Inserting this together with (9.11) into (9.12), we end up
with a candidate consumption function,
c
0
=
(1+v)[1(1+j)
10
(1+v)
(10)0
]
1(1+j)
T0
(1+v)
(10)T0
(c
0
+ /
0
) when

1+v
1+j
10
6= 1 + :.
1+v
T
(c
0
+ /
0
) when

1+v
1+j
10
= 1 + :.
(9.15)
For the subsequent periods we have from(9.11) that c
t
= ((1 + :),(1 + j))
t0
c
0
.
t = 1.. . . . 1 1.
EXAMPLE 1 Consider the special case o = 1 (i.e., n(c) = ln c) together
with j 0. The upper case in (9.15) is here the relevant one and period-0
consumption will be
c
0
=
(1 + :)(1 (1 + j)
1
)
1 (1 + j)
T
(c
0
+ /
0
) for o = 1.
We see that c
0
(1 + :)j(1 + j)
1
(c
0
+ /
0
) for 1 .
We have assumed that payment for consumption occurs at the end of the
period at the price 1 per consumption unit. To compare with the correspond-
ing result in continuous time with continuous compounding (see Section 9.4),
we might want to have initial consumption in the same present value terms
as c
0
and /
0
. That is, we consider c
0
c
0
(1 +:)
1
= j(1 +j)
1
(c
0
+/
0
).
That our candidate consumption function is indeed an optimal solution
when c
0
+ /
0
0 follows by concavity of the objective function (or by
concavity of the Hamiltonian if one applies the Maximum Principle of the
previous chapter). The conclusion is that under the idealized conditions
assumed, including a perfect loan market and perfect foresight, it is only
initial wealth and the interest rate that aect the time prole of consumption.
The time prole of income does not matter because consumption can be
smoothed over time by drawing on the bond market. Consumers look beyond
current income.
EXAMPLE 2 Consider the special case j = : 0. Again the upper case in
(9.15) is the relevant one and period-0 consumption will be
c
0
=
:
1 (1 + :)
T
(c
0
+ /
0
).
We see that c
0
:(c
0
+ /
0
) for 1 . So, with an innite time horizon
current consumption equals the interest on total current wealth. By con-
suming this the individual or household maintains total wealth intact. This
consumption function provides an interpretation of Milton Friedmans per-
manent income hypothesis. Friedman dened permanent income as the
amount a consumer unit could consume (or believes it could) while maintain-
ing its wealth intact (Friedman, 1957). The key point of Friedmans theory
was the idea that a random change in current income only aects current
consumption to the extent that it aects permanent income.
Alternative approach based on the intertemporal budget constraint
There is another approach to the households saving problem. With its choice
of consumption plan the household must act in conformity with its intertem-
poral budget constraint (IBC for short). The present value of the consump-
tion plan (c
1
. .... c
T1
), as seen from time zero, is
1\ (c
0
. c
1
. .... c
T1
)
T1
X
t=0
c
t
t
t=0
(1 + :
t
)
. (9.16)
This value cannot exceed the households total initial wealth, c
0
+/
0
. So the
households intertemporal budget constraint is
T1
X
t=0
c
t
t
t=0
(1 + :
t
)
c
0
+ /
0
. (9.17)
In this setting the households problem is to choose its consumption plan so
as to maximize l in (9.4) subject to this budget constraint.
This way of stating the problem is equivalent to the approach above
based on the dynamic budget condition (9.6) and the solvency condition
(9.7). Indeed, given the accounting equation (9.6), the consumption plan
of the household will satisfy the intertemporal budget constraint (9.17) if
and only if it satises the solvency condition (9.7). And there will be strict
equality in the intertemporal budget constraint if and only if there is strict
equality in the solvency condition (the proof is similar to that of a similar
claim relating to the government sector in Chapter 6.2).
338
Moreover, since in our specic saving problem saturation is impossible, an
optimal solution must imply strict equality in (9.17). So it is straightforward
to apply the substitution method also within the IBC approach. Alterna-
tively one can introduce the Lagrange function associated with the problem
of maximizing l =
P
T1
t=0
(1 + j)
t
n(c
t
) s.t. (9.17) with strict equality.
9.3 Transition to continuous time analysis
In the discrete time framework the run of time is divided into successive
periods of equal length, taken as the time-unit. Let us here index the periods
by i = 0. 1. 2. .... Thus nancial wealth accumulates according to
c
i+1
c
i
= :
i
. c
0
given,
where :
i
is (net) saving in period i.
Multiple compounding per year
With time owing continuously, we let c(t) refer to nancial wealth at time
t. Similarly, c(t +t) refers to nancial wealth at time t +t. To begin with,
let t equal one time unit. Then c(it) equals c(i) and is of the same value
as c
i
. Consider the forward rst dierence in c. c(t) c(t +t) c(t). It
makes sense to consider this change in c in relation to the length of the time
interval involved, that is, to consider the ratio c(t),t. As long as t = 1.
with t = it we have c(t),t = (c
i+1
c
i
),1 = c
i+1
c
i
. Now, keep
the time unit unchanged, but let the length of the time interval [t. t +t)
approach zero, i.e., let t 0. When c() is a dierentiable function, we
have
lim
t0
c(t)
t
= lim
t0
c(t +t) c(t)
t
=
dc(t)
dt
.
where dc(t),dt. often written c(t). is known as the derivative of c() at the
point t. Wealth accumulation in continuous time can then be written
c(t) = :(t). c(0) = c
0
given, (9.18)
where :(t) is the saving ow at time t. For t small we have the approx-
imation c(t) c(t)t = :(t)t. In particular, for t = 1 we have c(t)
= c(t + 1) c(t) :(t).
As time unit choose one year. Going back to discrete time we have that if
wealth grows at a constant rate q 0 per year, then after i periods of length
one year, with annual compounding, we have
c
i
= c
0
(1 + q)
i
. i = 0. 1. 2. ... . (9.19)
9.3. Transition to continuous time analysis 339
If instead compounding (adding saving to the principal) occurs : times a
year, then after i periods of length 1,: year and a growth rate of q,: per
such period,
c
i
= c
0
(1 +
q
:
)
i
. (9.20)
With t still denoting time measured in years passed since date 0, we have
i = :t periods. Substituting into (9.20) gives
c(t) = c
at
= c
0
(1 +
q
:
)
at
= c
0
(1 +
1
:
)
n
jt
. where :
:
q
.
We keep q and t xed, but let : and thus : . Then, in the limit
there is continuous compounding and it can be shown that
c(t) = c
0
c
jt
. (9.21)
where c is a mathematical constant called the base of the natural logarithm
and dened as c lim
n
(1 + 1,:)
n
' 2.7182818285....
The formula (9.21) is the continuous-time analogue to the discrete time
formula (9.19) with annual compounding. A geometric growth factor is re-
placed by an exponential growth factor.
We can also view the formulas (9.19) and (9.21) as the solutions to a
dierence equation and a dierential equation, respectively. Thus, (9.19) is
the solution to the linear dierence equation c
i+1
= (1+q)c
i
, given the initial
value c
0
. And (9.21) is the solution to the linear dierential equation c(t) =
qc(t). given the initial condition c(0) = c
0
. Now consider a time-dependent
growth rate, q(t). The corresponding dierential equation is c(t) = q(t)c(t)
and it has the solution
c(t) = c(0)c
I
0
j(t)ot
. (9.22)
where the exponent,
R
t
0
q(t)dt, is the denite integral of the function q(t)
from 0 to t. The result (9.22) is called the basic accumulation formula in
continuous time and the factor c
I
0
j(t)ot
is called the growth factor or the
accumulation factor.
7
Compound interest and discounting in continuous time
Let :(t) denote the short-term real interest rate in continuous time at time t.
To clarify what is meant by this, consider a deposit of \ (t) euro on a drawing
account in a bank at time t. If the general price level in the economy at time
7
Sometimes the accumulation factor with time-dependent growth rate is written in a
dierent way, see Appendix B.
340
t is 1(t) euro, the real value of the deposit is c(t) = \ (t),1(t) at time t.
By denition the real rate of return on the deposit in continuous time (with
continuous compounding) at time t is the (proportionate) instantaneous rate
at which the real value of the deposit expands per time unit when there is
no withdrawal from the account. Thus, if the instantaneous nominal interest
rate is i(t). we have

\ (t),\ (t) = i(t) and so, by the fraction rule in continuous
time (cf. Appendix A),
:(t) =
c(t)
c(t)
=

\ (t)
\ (t)

1(t)
1(t)
= i(t) :(t). (9.23)
where :(t)

1(t),1(t) is the instantaneous ination rate. In contrast to the
corresponding formula in discrete time, this formula is exact. Sometimes i(t)
and :(t) are referred to as the nominal and real interest intensity, respectively,
or the nominal and real force of interest.
Calculating the terminal value of the deposit at time t
1
t
0
. given its
value at time t
0
and assuming no withdrawal in the time interval [t
0
. t
1
], the
accumulation formula (9.22) immediately yields
c(t
1
) = c(t
0
)c
I
1
I
0
v(t)ot
.
When calculating present values in continuous time analysis, we use com-
pound discounting. We simply reverse the accumulation formula and go from
the compounded or terminal value to the present value c(t
0
). Similarly, given
a consumption plan, (c(t))
t
1
t=t
0
, the present value of this plan as seen from
time t
0
is
1\ =
Z
t
1
t
0
c(t) c
vt
dt. (9.24)
presupposing a constant interest rate. Instead of the geometric discount
factor, 1,(1 + :)
t
. from discrete time analysis, we have here an exponential
discount factor, 1,(c
vt
) = c
vt
. and instead of a sum, an integral. When the
interest rate varies over time, (9.24) is replaced by
1\ =
Z
t
1
t
0
c(t) c
I
I
0
v(t)ot
dt.
In (9.24) c(t) is discounted by c
vt
(1 +:)
t
for : small. This might
not seem analogue to the discrete-time discounting in (9.16) where it is c
t1
that is discounted by (1 + :)
t
. assuming a constant interest rate. When
taking into account the timing convention that payment for c
t1
in period
t 1 occurs at the end of the period (= time t). there is no discrepancy,
however, since the continuous-time analogue to this payment is c(t).
The allowed range for parameter values
The allowed range for parameters may change when we go from discrete time
to continuous time with continuous compounding. For example, the usual
equation for aggregate capital accumulation in continuous time is
1(t) = 1(t) o1(t). 1(0) = 1

0
given, (9.25)
where 1(t) is the capital stock, 1(t) is the gross investment at time t and
o 0 is the (physical) capital depreciation rate. Unlike in discrete time, here
o 1 is conceptually allowed. Indeed, suppose for simplicity that 1(t) = 0
for all t 0; then (9.25) gives 1(t) = 1
0
c
ct
. This formula is meaningful
for any o 0. Usually, the time unit used in continuous time macro models
is one year (or, in business cycle theory, rather a quarter of a year) and then
a realistic value of o is of course < 1 (say, between 0.05 and 0.10). However,
if the time unit applied to the model is large (think of a Diamond-style OLG
model), say 30 years, then o 1 may t better, empirically, if the model
is converted into continuous time with the same time unit. Suppose, for
example, that physical capital has a half-life of 10 years. With 30 years as
our time unit, inserting into the formula 1,2 = c
c3
gives o = (ln 2) 3 ' 2.
In many simple macromodels, where the level of aggregation is high, the
relative price of a unit of physical capital in terms of the consumption good
is 1 and thus constant. More generally, if we let the relative price of the
capital good in terms of the consumption good at time t be j(t) and allow
j(t) 6= 0. then we have to distinguish between the physical depreciation
of capital, o. and the economic depreciation, that is, the loss in economic
value of a machine per time unit. The economic depreciation will be d(t) =
j(t)o j(t). namely the economic value of the physical wear and tear (and
technological obsolescence, say) minus the capital gain (positive or negative)
on the machine.
Other variables and parameters that by denition are bounded frombelow
in discrete time analysis, but not so in continuous time analysis, include rates
of return and discount rates in general.
Stocks and ows
An advantage of continuous time analysis is that it forces the analyst to make
a clear distinction between stocks (say wealth) and ows (say consumption
or saving). Recall, a stock variable is a variable measured as a quantity at a
given point in time. The variables c(t) and 1(t) considered above are stock
variables. A ow variable is a variable measured as quantity per time unit
at a given point in time. The variables :(t).

1(t) and 1(t) are ow variables.
342
One can not add a stock and a ow, because they have dierent denomi-
nations. What exactly is meant by this? The elementary measurement units
in economics are quantity units (so many machines of a certain kind or so
many liters of oil or so many units of payment, for instance) and time units
(months, quarters, years). On the basis of these we can form composite mea-
surement units. Thus, the capital stock, 1. has the denomination quantity
of machines, whereas investment, 1. has the denomination quantity of ma-
chines per time unit or, shorter, quantity/time. A growth rate or interest
rate has the denomination (quantity/time)/quantity = time
1
. If we
change our time unit, say from quarters to years, the value of a ow variable
as well as a growth rate is changed, in this case quadrupled (presupposing
annual compounding).
In continuous time analysis expressions like 1(t)+1(t) or 1(t)+

1(t) are
thus illegitimate. But one can write 1(t +t) 1(t) +(1(t) o1(t))t. or
1(t)t (1(t) o1(t))t. In the same way, suppose a bath tub at time t
contains 50 liters of water and that the tap pours
1
2
liter per second into the
tub for some time. Then a sum like 50 / +
1
2
(//sec) does not make sense. But
the amount of water in the tub after one minute is meaningful. This amount
would be 50 / +
1
2
60 ((//sec)sec) = 80 /. In analogy, economic ow
variables in continuous time should be seen as intensities dened for every
t in the time interval considered, say the time interval [0, 1) or perhaps
[0, ). For example, when we say that 1(t) is investment at time t,
this is really a short-hand for investment intensity at time t. The actual
investment in a time interval [t
0
. t
0
+t) . i.e., the invested amount during
this time interval, is the integral,
R
t
0
+t
t
0
1(t)dt 1(t
0
)t. Similarly, the ow
of individual saving, :(t). should be interpreted as the saving intensity at
time t. The actual saving in a time interval [t
0
. t
0
+t) . i.e., the saved (or
accumulated) amount during this time interval, is the integral,
R
t
0
+t
t
0
:(t)dt.
If t is small, this integral is approximately equal to the product :(t
0
) t,
cf. the hatched area in Fig. 9.1.
The notation commonly used in discrete time analysis blurs the distinc-
tion between stocks and ows. Expressions like c
i+1
= c
i
+:
i
. without further
comment, are usual. Seemingly, here a stock, wealth, and a ow, saving, are
added. In fact, however, it is wealth at the beginning of period i and the
saved amount during period i that are added: c
i+1
= c
i
+ :
i
t. The tacit
condition is that the period length, t. is the time unit, so that t = 1.
But suppose that, for example in a business cycle model, the period length
is one quarter, but the time unit is one year. Then saving in quarter i is :
i
= (c
i+1
c
i
) 4 per year.

0
t
0
( ) s t
s
( ) s t
0
t
0
t t
Figure 9.1: With t small the integral of :(t) from t
0
to t
0
+t is the hatched
area.
The choice between discrete and continuous time analysis
In empirical economics, data typically come in discrete time form and data
for ow variables typically refer to periods of constant length. One could
argue that this discrete form of the data speaks for discrete time rather than
continuous time modelling. And the fact that economic actors often think
and plan in period terms, may seem a good reason for putting at least mi-
croeconomic analysis in period terms. Nonetheless real time is continuous.
And it can hardly be said that the mass of economic actors think and plan
with one and the same period. In macroeconomics we consider the sum of
the actions. In this perspective the continuous time approach has the advan-
tage of allowing variation within the usually articial periods in which the
data are chopped up. And for example centralized asset markets equilibrate
almost instantaneously and respond immediately to new information. For
such markets a formulation in continuous time seems preferable.
There is also a risk that a discrete time model may generate articial
oscillations over time. Suppose the true model of some mechanism is given
by the dierential equation
r = cr. c < 1. (9.26)
The solution is r(t) = r(0)c
ct
which converges in a monotonic way toward 0
for t . However, the analyst takes a discrete time approach and sets up
the seemingly corresponding discrete time model
r
t+1
r
t
= cr
t
.
344
This yields the dierence equation r
t+1
= (1 + c)r
t
, where 1 + c < 0. The
solutuion is r
t
= (1 + c)
t
r
0
. t = 0. 1. 2. . . . . As (1 + c)
t
is positive when t
is even and negative when t is odd, oscillations arise in spite of the true
model generating monotonous convergence towards the steady state r
= 0.
It should be added, however, that this potential problem can always be
avoided within discrete time models by choosing a suciently short period
length. Indeed, the solution to a dierential equation can always be ob-
tained as the limit of the solution to a corresponding dierence equation for
the period length approaching zero. In the case of (9.26) the approximating
dierence equation is r
i+1
= (1 + ct)r
i
. where t is the period length,
i = t,t, and r
i
= r(it). By choosing t small enough, the solution comes
arbitrarily close to the solution of (9.26). It is generally more dicult to go
in the opposite direction and nd a dierential equation that approximates
a given dierence equation. But the problem is solved as soon as a dier-
ential equation has been found that has the initial dierence equation as an
approximating dierence equation.
From the point of view of the economic contents, the choice between
discrete time and continuous time may be a matter of taste. From the point
of view of mathematical convenience, the continuous time formulation, which
has worked so well in the natural sciences, seems preferable. At least this is so
in the absence of uncertainty. For problems where uncertainty is important,
discrete time formulations are easier to work with unless one is familiar with
stochastic calculus.
9.4 Maximizing discounted utility in contin-
uous time
9.4.1 The saving problem in continuous time
In continuous time the analogue to the intertemporal utility function, (9.3),
is
l
0
=
R
T
0
n(c(t))c
jt
dt. (9.27)
In this context it is common to name the utility ow, n(). the instantaneous
utility function. We still assume that n
0
0 and n
00
< 0. The analogue to
the intertemporal budget constraint in Section 9.2 is
R
T
0
c(t)c
R
t
0
v(t)ot
dt c
0
+ /
0
. (9.28)
where, as above, c
0
is the historically given initial nancial wealth (the value
of the stock of short-term bonds held at time 0), while /
0
is the given human
9.4. Maximizing discounted utility in continuous time 345
wealth,
/
0
=
R
T
0
n(t)c
R
t
0
v(t)ot
dt. (9.29)
The households problem is then to choose a consumption plan (c(t))
T
I=0
so as to maximize discounted utility, l
0
. subject to the budget constraint
(9.28).
Innite time horizon In the Ramsey model of the next chapter the idea is
used that households may have an innite time horizon. One interpretation
of this is that parents care about their childrens future welfare and leave
bequests accordingly. This gives rise to a series of intergenerational links.
The household is then seen as a family dynasty with a time horizon far beyond
the lifetime of the current members of the family; Barros bequest model in
Chapter 7 is a discrete time application of this idea. Introducing a positive
constant utility discount rate, less weight is attached to circumstances further
away in the future and it may be ensured that the improper integral of
achievable discounted utilities over an innite horizon is bounded from above.
One could say, of course, that innity is a long time. The sun will even-
tually, in some billion years, burn out and life on earth become extinct.
Nonetheless an innite time horizon may provide a useful substitute for -
nite but remote horizons. This is because the solution to an optimization
problem for 1 large will in many cases in a large part of [0. 1] be close
to the solution for 1 .
8
And an innite time horizon may make ag-
gregation easier because at any future point in time, remaining time is still
innite. An innite time horizon can also sometimes be a convenient notion
when in any given period there is a positive probability that there will also be
a next period to be concerned about. If this probability is low, it can simply
be reected in a high eective utility discount rate. This idea is applied in
chapters 12 and 13.
We perform the transition to innite horizon by letting 1 in (9.27),
(9.28), and (9.29). In the limit the households, or dynastys, problem be-
comes one of choosing a plan, (c(t))
t=0
, which maximizes
l
0
=
Z

0
n(c(t))c
jt
dt s.t. (9.30)
Z

0
c(t)c
I
0
v(t)ot
dt c
0
+ /
0
. (IBC)
where /
0
emerges by letting 1 in (9.29) approach . Working with innite
horizons, there may exist technically feasible paths along which the improper
8
The turnpike proposition in Chapter 8 exemplies this.
346
integrals in (9.27), (9.28), and (9.29) go to for 1 . In that case
maximization is not well-dened. However, the assumptions that we are
going to make when working with the Ramsey model will guarantee that
the integrals converge as 1 (or at least that some feasible paths have
< l
0
< . while the remainder have l
0
= and are thus clearly
inferior). The essence of the matter is that the rate of time preference, j. must
be assumed suciently high relative to the potential growth in instantaneous
utility so as to ensure that the interest rate becomes higher than the long-run
growth rate of income.
Generally we dene a person as solvent if she is able to meet her nancial
obligations as they fall due. Each person is considered small relative to the
economy as a whole. As long as all agents in the economy remain small,
they will in general equilibrium remain solvent if and only if their gross debt
does not exceed their gross assets. Because of our assumption of a perfect loan
market, assets should here be understood in the broadest possible sense,
that is, including the present value of the expected future labor income.
Considering net debt, d
0
, the solvency requirement becomes
d
0

Z

0
(n(t) c(t))c
I
0
v(t)ot
dt.
where the right-hand side of the inequality is the present value of the expected
future primary saving.
9
By use of the denition in (9.29), it can be seen that
this requirement is identical to the intertemporal budget constraint (IBC)
which consequently expresses solvency.
The budget constraint in ow terms
The method which is particularly apt for solving intertemporal decision prob-
lems in continuous time is based on the mathematical discipline optimal
control theory. To apply the method, we have to convert the households
budget constraint from the present value formulation considered above into
ow terms.
By mere accounting, in every short time interval (t, t+t) the households
consumption plus saving equals the households total income, that is,
(c(t) + c(t))t = (:(t)c(t) + n(t))t.
Here, c(t) dc(t),dt is saving and thus the same as the increase per time
unit in nancial wealth. If we divide through by t and isolate saving on
9
By primary saving is meant the dierence between current earned income and current
consumption, where earned income means income before interest transfers.
the left-hand side of the equation, we get for all t 0
c(t) = :(t)c(t) + n(t) c(t). c(0) = c
0
given. (9.31)
This equation in itself is just a dynamic budget identity. It tells us by how
much and in which direction the nancial wealth is changing due to the dif-
ference between current income and current consumption. The equation per
se does not impose any restriction on consumption over time. If this equa-
tion were the only restriction, one could increase consumption indenitely
by incurring an increasing debt without limits. It is not until we add the
requirement of solvency that we get a constraint. When 1 < , the relevant
solvency requirement is c(1) 0 (that is, no debt left over at the terminal
date). This is equivalent to satisfying the intertemporal budget constraint
(9.28). When 1 = . the relevant solvency requirement is a No-Ponzi-Game
condition:
lim
t
c(t)c
I
0
v(t)ot
0. (NPG)
i.e., the present value of debts, measured as c(t). innitely far out in the
future, is not permitted to be positive. Indeed, we have the following equiv-
alency:
PROPOSITION 1 (equivalence of NPG condition and intertemporal budget
constraint) Let the time horizon be innite and assume that the integral
(9.29) remains nite for 1 . Then, given the accounting relation (9.31),
we have:
(i) the requirement (NPG) is satised if and only if the intertemporal budget
constraint, (IBC), is satised; and
(ii) there is strict equality in (NPG) if and only if there is strict equality in
(IBC).
Proof. See Appendix C.
The condition (NPG) does not preclude that the household (or rather
the family dynasty) can remain in debt. This would also be an unnatural
requirement as the dynasty is innitely-lived. The condition does imply,
however, that there is an upper bound for the speed whereby debts can
increase in the long term. The NPG condition says that in the long term,
debts are not allowed to grow at a rate greater than or equal to the interest
rate.
To understand the implication, let us look at the case where the interest
rate is a constant, : 0. Assume that the household at time t has net
debt d(t) 0, i.e., c(t) d(t) < 0. If d(t) were persistently growing at a
348
rate equal to or greater than the interest rate, (NPG) would be violated.
10
Equivalently, one can interpret (NPG) as an assertion that lenders will only
issue loans if the borrowers in the long run are able to cover at least part of
their interest payments by other means than by taking up new loans. In this
way, it is avoided that

d(t) :d(t) in the long run, that is, the debt does
not explode.
As mentioned in Chapter 6 the name No-Ponzi-Game condition refers
to a guy, Charles Ponzi, who in Boston in the 1920s temporarily became very
rich by a loan arrangement based on the chain letter principle. The fact that
debts grow without bounds is irrelevant for the lender if the borrower can
always nd new lenders and use their outlay to pay o old lenders. In the
real world, endeavours to establish this sort of nancial eternity machine tend
sooner or later to break down because the ow of new lenders dries up. Such
nancial arrangements, in everyday speech known as pyramid companies, are
universally illegal.
11
It is exactly such arrangements the constraint (NPG)
precludes.
9.4.2 Solving the saving problem
The households consumption/saving problem is one of choosing a path for
the control variable c(t) so as to maximize a criterion function, in the form
of an integral, subject to constraints that include a rst-order dierential
equation. This equation determines the evolution of the state variable, c(t).
Optimal control theory, which in Chapter 8 was applied to a related discrete
time problem, oers a well-suited apparatus for solving this kind of optimiza-
tion problem. We will make use of a special case of Pontryagins Maximum
Principle (the basic tool of optimal control theory) in its continuous time
version. We shall consider the case with a nite as well as innite time
horizon.
10
Starting from a given initial positive debt, d
0
, when

d(t),d(t) r 0, we have
d(t) d
0
c
so that d(t)c
d
0
0 for all t 0. Consequently, a(t)c
= d(t)c
d
0
< 0 for all t 0, that is, lim

a(t)c
< 0, which violates (NPG).

11
A related Danish instance, though on a modest scale, could be read in the Danish
newpaper Politiken on the 21st of August 1992. A twenty-year-old female student from
Tylstrup in Northern Jutland is charged with fraud. In an ad, she oered 200 DKK to tell
you how to make easy money. Some hundred people responded and received the reply:
Do like me.
A more serious present day example is the American stockbroker, Bernard Mado, who
admitted a Ponzi scheme that is considered to be the largest nancial fraud in U.S. history.
In 2009 Mado was sentenced to 150 years in prison.
For 1 < the problem is: choose a plan (c(t))
T
t=0
that maximizes
l
0
=
Z
T
0
n(c(t))c
jt
dt s.t. (9.32)
c(t) 0. (control region) (9.33)
c(t) = :(t)c(t) + n(t) c(t). c(0) = c
0
given, (9.34)
c(1) 0. (9.35)
With an innite time horizon, 1 in (9.32) is interpreted as and the solvency
condition (9.35) is replaced by
lim
t
c(t)c
I
0
v(t)ot
0. (NPG)
Let 1 denote the time interval [0. 1] if 1 < and the time interval
[0. ) if 1 = . If c(t) and the corresponding evolution of c(t) full (9.33)
and (9.34) for all t 1 as well as the relevant solvency condition, we call
(c(t). c(t))
T
t=0
an admissible path. If a given admissible path (c(t). c(t))
T
t=0
solves the problem, it is referred to as an optimal path.
12
We assume that
n(t) and :(t) are piecewise continuous functions of t and that n(t) is positive
for all t. No condition on the impatience parameter, j. is imposed (in this
chapter).
First-order conditions
The solution procedure for this problem is as follows:
13
1. We set up the current-value Hamiltonian function:
H(c. c. `. t) n(c) + `(:c + n c).
where ` is the adjoint variable (also called the co-state variable) asso-
ciated with the dynamic constraint (9.34).
14
That is, ` is an auxiliary
variable which is a function of t and is analogous to the Lagrange mul-
tiplier in static optimization.
12
The term path, sometimes trajectory, is common in the natural sciences for a
solution to a dierential equation because one may think of this solution as the path of a
particle moving in two- or three-dimensional space.
13
The four-step solution procedure below is applicable to a large class of dynamic opti-
mization problems in continuous time, see Math tools.
14
The explicit dating of the time-dependent variables a, c, and ` is omitted where not
needed for clarity.
350
2. At every point in time, we maximize the Hamiltonian w.r.t. the control
variable, in the present case c. So, focusing on an interior optimal
path,
15
we calculate
JH
Jc
= n
0
(c) ` = 0.
For every t 1 we thus have the condition
n
0
(c(t)) = `(t). (9.36)
3. We calculate the partial derivative of H with respect to the state vari-
able, in the present case c, and put it equal to the dierence between the
discount rate (as it appears in the integrand of the criterion function)
multiplied by ` and the time derivative of ` :
JH
Jc
= `: = j`

`.
That is, for all t 1, the adjoint variable ` should full the dierential
equation
`(t) = (j :(t))`(t). (9.37)

4. We now apply the Maximum Principle which applied to this problem
says: an interior optimal path (c(t). c(t))
T
t=0
will satisfy that there exits
a continuous function ` = `(t) such that for all t 1. (9.36) and (9.37)
hold along the path, and the transversality condition,
c(1)`(1)c
jT
= 0. if 1 < . or
lim
t
c(t)`(t)c
jt
= 0. if 1 = . (TVC)
is satised.
An optimal path is thus characterized as a path that for every t maxi-
mizes the Hamiltonian associated with the problem. The intuition is that
the Hamiltonian weighs the direct contribution of the marginal unit of the
control variable to the criterion function in the right way relative to the
indirect contribution, which comes from the generated change in the state
variable (here nancial wealth); right means in accordance with the op-
portunities oered by the rate of return vis-a-vis the time preference rate, j.
The optimality condition (9.36) can be seen as a `C = `1 condition: on
the margin one unit of account (here the consumption good) must be equally
valuable in its two uses: consumption and wealth accumulation. Together
15
A path, (a
, c
=0
, is an interior path if c
0 for all t 0.
with the optimality condition (9.37) this signies that the adjoint variable `
can be interpreted as the shadow price (measured in units of current utility)
of nancial wealth along the optimal path.
16
Remark. The current-value Hamiltonian function is often just called the
current-value Hamiltonian. More importantly the prex current-value is
used to distinguish it from what is known as the present-value Hamiltonian.
The latter is dened as

H Hc
jt
with `c
jt
substituted by j. which is the
associated (discounted) adjoint variable. The solution procedure is similar
except that step 3 is replaced by J

H,Jc = j and `(t)c
jt
in the transver-
sality condition is replaced by j(t). The two methods are equivalent. But for
many economic problems the current-value Hamiltonian has the advantage
that it makes both the calculations and the interpretation slightly simpler.
The adjoint variable, `(t), which as mentioned acts as a shadow price of the
state variable, is a current price along with the other prices in the problem,
n(t) and :(t). in contrast to j(t) which is a discounted price.
Reordering (9.37) gives
:` +

`
`
= j. (9.38)
This can be interpreted as a no-arbitrage condition. The left-hand side gives
the actual rate of return, measured in utility units, on the marginal unit of
saving. Indeed, :` can be seen as a dividend and

` as a capital gain. The-
right hand side is the required rate of return measured in utility terms, j.
Along an optimal path the two must coincide. The household is willing to
save the marginal unit only up to the point where the actual rate of return
on saving equals the required rate.
We may alternatively write the no-arbitrage condition as
: = j

`
`
. (9.39)
On the left-hand-side appears the actual real rate of return on saving and
on the right-hand-side the required real rate of return. The intuition behind
this condition can be seen in the following way. Suppose Ms. Jones makes a
deposit of \ utility units in a bank that oers a proportionate rate of ex-
pansion of the utility value of the deposit equal to i (assuming no withdrawal
occurs), i.e.,
\
\
= i.
16
Recall, a shadow price (measured in some unit of account) of a good is the number of
units of account that the optimizing agent is just willing to oer for one extra unit of the
good.
352
This is the actual utility rate of return, a kind of nominal interest rate. To
calculate the corresponding real interest rate let the nominal price of a
consumption good be ` utility units. Dividing the number of invested utility
units, \. by `. we get the real value, : = \,`. of the deposit at time t. The
actual real rate of return on the deposit is therefore
: =
:
:
=

\
\

`
`
= i

`
`
. (9.40)
Ms. Jones is then just willing to save the marginal unit of income if this real
rate of return on saving equals the required real rate, that is, the right-hand
side of (9.39); in turn this necessitates that the nominal interest rate i
equals the required nominal rate, j. The formula (9.40) is analogue to the
discrete-time formula (9.2) except that the unit of account in (9.40) is current
utility while in (9.2) it is currency.
Substituting (9.36) into the transversality condition for the case 1 < .
gives
c(1)c
jT
n
0
(c(1)) = 0. (9.41)
Our solvency condition, c(1) 0. can be seen as an example of a general
inequality constraint, c(1) c
T
. where here c
T
happens to equal 0. So
(9.41) can be read as a standard complementary slackness condition. The
(discounted) price, c
jT
n
0
(c(1)). is always positive, hence an optimal plan
must satisfy c(1) = c
T
(= 0). The alternative, c(1) 0. would imply
that consumption, and thereby l
0
, could be increased by a decrease in c(1)
without violating the solvency requirement.
Now let 1 . Then in the limit the solvency requirement is (NPG),
and (9.41) is replaced by
lim
T
c(1)c
jT
n
0
(c(1)) = 0. (9.42)
This is the same as (TVC) (replace 1 by t). Intuitively, a plan that violates
this condition by having instead = indicates scope for improvement
and thus cannot be optimal. There would be purchasing power left for
eternity which could be transferred to consumption on earth at an earlier
date.
Generally, care must be taken when extending a necessary transversality
condition from a nite to an innite horizon. But for the present problem,
the extension is valid. Indeed, (TVC) is just a requirement that the NPG
condition is not over-satised:
PROPOSITION2 (the households transversality condition with innite time
horizon) Let 1 and assume the integral (9.29) remains nite for 1
. Provided the adjoint variable, `(t), satises the optimality conditions
(9.36) and (9.37), (TVC) holds if and only if (NPG) holds with strict equality.
Proof. See Appendix D.
In view of this proposition, we can write the transversality condition for
1 as the NPG condition with strict equality:
lim
t
c(t)c
I
0
v(t)ot
= 0. (TVC)
9.4.3 The Keynes-Ramsey rule
The rst-order conditions have interesting implications. Dierentiate both
sides of (9.36) w.r.t. t to get n
00
(c) c =

` which can be written as n
00
(c) c,n
0
(c) =
`,` by drawing on (9.36) again. Applying (9.37) now gives

c(t)
c(t)
=
1
o(c(t))
(:(t) j). (9.43)
where o(c) is the (absolute) elasticity of marginal utility w.r.t. consumption,
o(c)
c
n
0
(c)
n
00
(c) 0. (9.44)
As in discrete time, o(c) indicates the strength of the consumers desire to
smooth consumption. The inverse of o(c) measures the instantaneous in-
tertemporal elasticity of substitution in consumption, which in turn indicates
the willingness to accept variation in consumption over time when the inter-
est rate changes, see Appendix F.
The result (9.43) says that an optimal consumption plan is characterized
in the following way. The household will completely smooth consumption
over time if the rate of time preference equals the real interest rate. The
household will choose an upward-sloping time path for consumption if and
only if the rate of time preference is less than the real interest rate. Indeed,
in this case the household would accept a relatively low level of current
consumption with the purpose of enjoying more consumption in the future.
The lower the rate of time preference relative to the real interest rate, the
more favorable it becomes to defer consumption. Moreover, by (9.43) we
see that the greater the elasticity of marginal utility (that is, the greater
the curvature of the utility function), the greater the incentive to smooth
consumption for a given value of :(t) j. The reason for this is that a large
curvature means that the marginal utility will drop sharply if consumption
increases, and will rise sharply if consumption falls. Fig. 9.2 illustrates this
354
t
lnc
0
1

2 1

Figure 9.2: Optimal consumption paths for a low and a high constant 0, given a
constant r j.
in the CRRA case where o(c) = o. a positive constant. For a given constant
: j. the consumption path chosen when o is high has lower slope, but
starts from a higher level, than when o is low.
The condition (9.43), which holds for a nite as well as an innite time
horizon, is referred to as the Keynes-Ramsey rule. The name springs from
the English mathematician Frank Ramsey who derived the rule in as early
as 1928, while John Maynard Keynes suggested a simple and intuitive way of
presenting it. The rule is the continuous-time counterpart to the consumption
Euler equation in discrete time.
The Keynes-Ramsey rule reects the general microeconomic principle
that the consumer equates the marginal rate of substitution between any
two goods with the corresponding price ratio. In the present context the
principle is applied to a situation where the two goods refer to the same
consumption good delivered at two dierent dates. In Section 9.2 we used
the principle to interpret the optimal saving behavior in discrete time. How
can the principle be translated into a continuous time setting?
Local optimality in continuous time Let (t, t+t) and (t+t, t+2t)
be two short successive time intervals. The marginal rate of substitution,
`1o
t+t,t
, of consumption in the second time interval for consumption in
the rst is
17
`1o
t+t,t

dc(t +t)
dc(t)
|
l=
l
=
n
0
(c(t))
c
jt
n
0
(c(t +t))
. (9.45)
17
The underlying analytical steps can be found in Appendix E.
approximately. On the other hand, by saving c(t) more per time unit
(where c(t) < 0) in the short time interval (t, t + t), one can via the
market transform c(t) t units of consumption in this time interval into
c(t +t) t c(t)t c
I+I
I
v(t)ot
(9.46)
units of consumption in the time interval (t + t, t + 2t). The marginal
rate of transformation is therefore
`11
t+t,t

dc(t +t)
dc(t)
|
l=
l

= c
I+I
I
v(t)ot
.
In the optimal plan we must have `1o
t+t,t
= `11
t+t,t
which gives
n
0
(c(t))
c
jt
n
0
(c(t +t))
= c
I+I
I
v(t)ot
. (9.47)
approximately. When t = 1 and j and :(t) are small, this relation can be
approximated by (9.9) from discrete time (generally, by a rst-order Taylor
approximation, c
a
1 + r. when r is close to 0).
Taking logs on both sides of (9.47), dividing through by t. inserting
(9.46), and letting t 0. we get (see Appendix E)
j
n
00
(c(t))
n
0
(c(t))
c(t) = :(t). (9.48)
With the denition of o(c) in (9.44), this is exactly the same as the Keynes-
Ramsey rule (9.43) which, therefore, is merely an expression of the general
optimality condition `1o = `11. When c(t) 0. the household is willing
to sacrice some consumption today for more consumption tomorrow only
if it is compensated by an interest rate suciently above j. Naturally, the
required compensation is higher, the faster marginal utility declines with
rising consumption, i.e., the larger is (n
00
,n
0
) c already. Indeed, a higher
c
t
in the future than today implies a lower marginal utility of consumption
in the future than of consumption today. So saving of the marginal unit of
income today is only warranted if the rate of return is suciently above j.
and this is what (9.48) indicates.
9.4.4 Mangasarians sucient conditions
The Maximum Principle delivers a set of rst-order and transversality condi-
tions that as such are only necessary conditions for an interior path to be op-
timal. Hence, up to this point we have only claimed that if the consumption-
saving problem has an interior solution, then it satises the Keynes-Ramsey
356
rule and a transversality condition, (TVC). But are these conditions also
sucient? The answer is yes in the present case. This follows from Man-
gasarians suciency theorem (see Math tools) which applied to the present
problem tells us that if the Hamiltonian is concave in (c. c) for every t, then
the listed necessary conditions, including the transversality condition, are
also sucient. Because the instantaneous utility function, the rst term in
the Hamiltonian, is strictly concave and the second term is linear in (c. c),
the Hamiltonian is concave in (c. c). Thus if we have found a path satisfying
the Keynes-Ramsey rule and the (TVC), we have a candidate solution. And
by the Mangasarian theorem this candidate is an optimal solution. In fact
the strict concavity of the Hamiltonian with respect to the control variable
ensures that the optimal solution is unique (Exercise 9.?).
9.5 The consumption function
We have not yet fully solved the saving problem. The Keynes-Ramsey rule
gives only the optimal rate of change of consumption over time. It says
nothing about the level of consumption at any given time. In order to de-
termine, for instance, the level c(0), we implicate the solvency condition
which limits the amount the household can borrow in the long term. Among
the innitely many consumption paths satisfying the Keynes-Ramsey rule,
the household will choose the highest one that also fulls the solvency re-
quirement (NPG). Thus, the household acts so that strict equality in (NPG)
obtains. As we saw, this is equivalent to the transversality condition being
satised.
To avoid any misunderstanding, the examples below should not be inter-
preted such that for any evolution of wages and interest rates there exists a so-
lution to the households maximization problem with innite horizon. There
is generally no guarantee that integrals have an upper bound for 1 . The
evolution of wages and interest rates which prevails in general equilibrium is
not arbitrary, however. It is determined by the requirement of equilibrium.
In turn, of course existence of an equilibrium imposes restrictions on the
utility discount rate relative to the potential growth in instantaneous utility.
We shall return to these issues in the next chapter.
EXAMPLE 1 (constant elasticity of marginal utility; innite time horizon).
In the problem in Section 9.4.2 with 1 = . we consider the case where the
elasticity of marginal utility o(c). as dened in (9.44), is a constant o 0.
From Appendix A of Chapter 3 we know that this requirement implies that
up to a positive linear transformation the utility function must be of the
9.5. The consumption function 357
form:
n(c) =

c
10
10
. when o 0. o 6= 1.
ln c. when o = 1.
(9.49)
The Keynes-Ramsey rule then implies c(t) = o
1
(:(t) j)c(t). Solving this
linear dierential equation yields
c(t) = c(0)c
1
0
I
0
(v(t)j)ot
. (9.50)
cf. the general accumulation formula, (9.22).
We know from Proposition 2 that the transversality condition is equiva-
lent to the intertemporal budget constraint being satised with strict equal-
ity, i.e.,
Z

0
c(t)c
I
0
v(t)ot
dt = c
0
+ /
0
. (IBC)
where /
0
is the human wealth,
/
0
=
Z

0
n(t)c
R
t
0
v(t)ot
dt. (9.51)
This result can be used to determine c(0).
18
Substituting (9.50) into (IBC)
gives
c(0)
Z

0
c
I
0
[
1
0
(v(t)j)v(t)]ot
dt = c
0
+ /
0
.
The consumption function is thus
c(0) = ,
0
(c
0
+ /
0
). where
,
0

1
R
0
c
I
0
[
1
0
(v(t)j)v(t)]ot
dt
=
1
R
0
c
1
0
I
0
[(10)v(t)j]ot
dt
(9.52)
is the marginal propensity to consume out of wealth. We have here assumed
that these improper integrals over an innite horizon are bounded from above
for all admissible paths.
Generally, an increase in the interest rate level, for given total wealth,
c
0
+ /
0
, can eect c(0) both positively and negatively.
19
On the one hand,
such an increase makes future consumption cheaper in present value terms.
This change in the trade-o between current and future consumption entails
18
The method also applies if instead of T = , we have T < .
19
By an increase in the interest rate level we mean an upward shift in the time-prole
of the interest rate. That is, there is at least one time interval within [0, ) where the
interest rate is higher than in the original situation and no time interval within [0, )
where the interest rate is lower.
358
a negative substitution eect on c(0). On the other hand, the increase in
the interest rates decreases the present value of a given consumption plan,
allowing for higher consumption both today and in the future, for given total
wealth, cf. (IBC). This entails a positive pure income eect on consumption
today as consumption is a normal good. If o < 1 (small curvature of the
utility function), the substitution eect will dominate the pure income eect,
and if o 1 (large curvature), the reverse will hold. This is because the larger
is o, the stronger is the propensity to smooth consumption over time.
In the intermediate case o = 1 (the logarithmic case) we get from (9.52)
that ,
0
= j, hence
c(0) = j(c
0
+ /
0
). (9.53)
In this special case the marginal propensity to consume is time independent
and equal to the rate of time preference. For a given total wealth, c
0
+
/
0
, current consumption is thus independent of the expected path of the
interest rate. That is, in the logarithmic case the substitution and pure
income eects on current consumption exactly oset each other. Yet, on top
of this comes the negative wealth eect on current consumption of an increase
in the interest rate level. The present value of future wage incomes becomes
lower (similarly with expected future dividends on shares and future rents
in the housing market in a more general setup). Because of this, /
0
(and
so c
0
+ /
0
) becomes lower, which adds to the negative substitution eect.
20
Thus, even in the logarithmic case, and a fortiori when o < 1. the total eect
of an increase in the interest rate level is unambiguously negative on c(0).
If, for example, :(t) = : and n(t) = n (positive constants), we get ,
0
= [(o 1): + j],o and c
0
+ /
0
= c
0
+ n,:. When o = 1. the negative eect
of a higher : on /
0
is decisive. When o < 1. a higher : reduces both ,
0
and
/
0
. hence the total eect on c(0) is even more negative. When o 1, a
higher : implies a higher ,
0
which more or less osets the lower /
0
, so that
the total eect on c(0) becomes ambiguous. As referred to in Chapter 3,
available empirical studies generally suggest a value of o somewhat above 1.
EXAMPLE 2 (constant absolute semi-elasticity of marginal utility; innite

time horizon). In the problem in Section 9.4.2 with 1 = . we consider the
case where the sensitivity of marginal utility, measured by the absolute value
of the semi-elasticity of marginal utility, n
00
(c),n
0
(c) (n
0
,n
0
),c, is a
positive constant, c. The utility function must then, up to a positive linear
20
If a
0
< 0 and this net debt is not a variable-rate loan (as hitherto assumed), but like
a xed-rate mortgage loan, then a rise in the interest rate level implies a lowering of the
present value of the debt and thereby raises total wealth (ceteris paribus) and counteracts
the negative substitution eect on current consumption.
9.5. The consumption function 359
transformation, be of the form,
n(c) = c
1
c
cc
. c 0. (9.54)
The Keynes-Ramsey rule becomes c(t) = c
1
(:(t) j). When the interest
rate is a constant : 0, we nd, through (IBC) and partial integration, c(0)
= :(c
0
+/
0
) (: j),(c:), presupposing : j and c
0
+/
0
(: j),(c:
2
).
This hypothesis of a constant absolute variability aversion implies that
the degree of relative variability aversion is o(c) = cc and thus greater, the
larger is c. In the theory of behavior under uncertainty, (9.54) is referred to
as the CARA function (Constant Absolute Risk Aversion). One of the the-
orems of expected utility theory is that the degree of absolute risk aversion,
n
00
(c),n
0
(c), is proportional to the risk premium which the economic agent
will require to be willing to exchange a specied amount of consumption re-
ceived with certainty for an uncertain amount having the same mean value.
Empirically this risk premium seems to be a decreasing function of the level
of consumption. Therefore the CARA function is generally considered less
realistic than the CRRA function of the previous example.
EXAMPLE 3 (logarithmic utility; nite time horizon; retirement). We con-
sider a life-cycle saving problem. A worker enters the labor market at time 0
with a nancial wealth of 0, has nite lifetime 1 (assumed known) and does
not wish to pass on bequests. For simplicity, we assume that :
t
= : 0 for
all t [0. 1] and n(t) = n 0 for t t
1
1, while n(t) = 0 for t t
1
(no
wage income after retirement, which takes place at time t
1
). The decision
problem is
max
(c(t))
T
I=0
l
0
=
Z
T
0
(lnc(t))c
jt
dt s.t.
c(t) 0.
c(t) = :c(t) + n(t) c(t). c(0) = 0.
c(1) 0.
The Keynes-Ramsey rule becomes c
t
,c
t
= :j. Asolution to the problem
will thus full
c(t) = c(0)c
(vj)t
. (9.55)
Inserting this into the dierential equation for c. we get a rst-order linear
dierential equation the solution of which (for c(0) = 0) can be reduced to
c(t) = c
vt
n
:
(1 c
v:
)
c
0
j
(1 c
jt
)
. (9.56)
360
where . = t if t t
1
, and . = t
1
if t t
1
. We need to determine c(0).
The transversality condition implies c(1) = 0. Having t = 1, . = t
1
and
c
T
= 0 in (9.56), we get
c(0) = (jn,:)(1 c
vt
1
),(1 c
jT
). (9.57)
Substituting this into (9.55) gives the optimal consumption plan.
21
If : = j, consumption is constant over time at the level given by (9.57).
If, in addition, t
1
< 1, this consumption level is less than the wage income
per year up to t
1
(in order to save for retirement); in the last years the level
of consumption is maintained although there is no wage income; the retired
person uses up both the return on nancial wealth and this wealth itself.
The examples illustrate the importance of forward-looking expectations,
here the expected evolution of interest rates and wages. The expectations
aect c(0) both through their impact on the marginal propensity to consume
(cf. ,
0
in Example 1) and through their impact on the present value, /
0
, of
expected future labor income (or of expected future dividends on shares in a
more general setup).
22
Yet the examples and the consumption theory in
this chapter in general should only be seen as a rst, crude approximation
to consumption/saving behavior. Real world factors such as uncertainty
and narrow credit constraints (absence of perfect loan markets) also aect
the behavior. Including these factors in the analysis tend to make current
income an additional determinant of the consumption by a large fraction
of the population, as is recognized in many short- and medium-run macro
models.
9.6 Literature notes
(incomplete)
Loewenstein and Thaler (1989) survey the evidence suggesting that the
utility discount rate is generally not constant, but declining with the time
distance from the current period to the future periods within the horizon.
This is known as hyperbolic discounting.
The (strong) assumptions regarding the underlying intertemporal pref-
erences which allow them to be represented by the present value of period
21
For t
1
= T and T we get in the limit c(0) = jn,r j/
0
, which is also what
(9.52) gives when a(0) = 0 and 0 = 1.
22
How to treat cases where, due to new information, a shift in expectations occurs so
that a discontinuity in a responding endogenous variable results is dealt with in Chapter
11.
9.7. Appendix 361
utilities discounted at a constant rate are dealt with by Koopmans (1960),
Fishburn and Rubinstein (1982), and in summary form by Heal (1998).
Rigorous and more general treatments of the Maximum Principle in con-
tinuous time applied in economic analysis are available in, e.g., Seierstad and
Sydsaeter (1987) and Sydsaeter et al. (2008).
Allen (1967). Goldberg (1958).
9.7 Appendix
A. Growth arithmetic in continuous time
Let the variables .. r. and be dierentiable functions of time t. Suppose
.(t). r(t). and (t) are positive for all t. Then:
PRODUCT RULE .(t) = r(t)(t)
:(t)
:(t)
=
a(t)
a(t)
+
j(t)
j(t)
.
Proof. Taking logs on both sides of the equation .(t) = r(t)(t) gives
ln .(t) = lnr(t) + ln (t). Dierentiation w.r.t. t, using the chain rule, gives
the conclusion.
The procedure applied in this proof is called logarithmic dierentiation
w.r.t. t.
FRACTION RULE .(t) =
a(t)
j(t)

:(t)
:(t)
=
a(t)
a(t)

j(t)
j(t)
.
The proof is similar.
POWER FUNCTION RULE .(t) = r(t)
c
:(t)
:(t)
= c
a(t)
a(t)
.
The proof is similar.
In continuous time these simple formulas are exactly true. In discrete time
the analogue formulas are only approximately true and the approximation
can be quite bad unless the growth rates of r and are small, cf. Appendix
A to Chapter 4.
B. The cumulative mean of growth and interest rates
Sometimes in the literature the basic accumulation formula, (9.22), is ex-
pressed in terms of the arithmetic average (also called the cumulative mean)
of the growth rates in the time interval [0. t]. This average is q
0,t
= (1,t)
R
t
0
q(t)dt.
So we can write
c(t) = c(0)c
j
0,I
t
.
which has form similar to (9.21). Similarly, let :
0,t
denote the arithmetic
average of the (short-term) interest rates from time 0 to time t. i.e., :
0,t
362
= (1,t)
R
t
0
:(t)dt. Then we can write the present value of the consumption
stream (c(t))
T
t=0
as 1\ =
R
T
0
c(t)c
v
0,I
t
dt.
In discrete time the arithmetic average of growth rates can at best be
used as an approximation. Let q
0,t
be the the average compound growth rate
from year 0 to year t. that is, 1 + q
0,t
= ((1 + q
0
)(1 + q
1
) (1 + q
t1
))
1t
. If
the period length is short, however, say a quarter of a year, then the growth
rates q
1
. q
2
...., hence also q
0,t
. will generally be not far from zero so that the
approximation ln(1+q
t
) q
t
is acceptable. Then q
0,t
1
t
(q
0
+q
1
+... +q
t1
).
a simple arithmetic average. As compounding is left out, this approximation
is not good if there are many periods unless the growth rates are very small
numbers.
Similarly with interest rates in discrete time.
C. Notes on Proposition 1 (equivalence between the No-Ponzi-
Game condition and the intertemporal budget constraint)
We consider the book-keeping relation
c(t) = :(t)c(t) + n(t) c(t). (9.58)
where c(0) = c
0
(given), and the solvency requirement
lim
t
c(t)c
I
0
v(t)ot
0. (NPG)
Technical remark. The expression in (NPG) is understood to include the
possibility that c(t)c
I
0
v(t)ot
for t . Moreover, if full generality
were aimed at, we should allow for innitely uctuating paths in both the
(NPG) and (TVC) and therefore replace lim
t
by liminf
t
, i.e., the
limit inferior. The limit inferior for t of a function ,(t) on [0. ) is
dened as lim
t
inf {,(:)| : t}.
23
As noted in Appendix E of the previous
chapter, however, undamped innitely uctuating paths never turn up in
the optimization problems considered in this book, whether in discrete or
continuous time. Hence, we apply the simpler concept lim rather than
liminf.
On the background of (9.58) Proposition 1 claimed that (NPG) is equiv-
alent with the intertemporal budget constraint,
Z
0
c(t)c
I
0
v(t)ot
dt /
0
+ c
0
. (IBC)
23
By inf is meant innum of the set, that is, the largest number less than or equal to
all numbers in the set.
9.7. Appendix 363
being satised, where /
0
is dened as in (9.51) and is assumed to be a nite
number. In addition, Proposition 1 in Section 9.4 claimed that there is strict
equality in (IBC) if and only there is strict equality in (NPG). We now prove
these claims.
Proof. Isolate c(t) in (9.58) and multiply through by c
R
t
0
v(t)ot
to obtain
c(t)c
R
t
0
v(t)ot
= n(t)c
R
t
0
v(t)ot
( c(t) :(t)c(t))c
R
t
0
v(t)ot
.
Integrate from 0 to 1 0 to get
R
T
0
c(t)c
R
t
0
v(t)ot
dt
=
Z
T
0
n(t)c
R
t
0
v(t)ot
dt
Z
T
0
c(t)c
R
t
0
v(t)ot
dt +
Z
T
0
:(t)c(t)c
R
t
0
v(t)ot
dt
=
Z
T
0
n(t)c
R
t
0
v(t)ot
dt
c(t)c
R
t
0
v(t)ot
T
0
Z
T
0
c(t)c
R
t
0
v(t)ot
(:(t))dt
!
+
Z
T
0
:(t)c(t)c
R
t
0
v(t)ot
dt
=
Z
T
0
n(t)c
R
t
0
v(t)ot
dt (c(1)c
T
0
v(t)ot
c(0)).
where the second last equality follows from integration by parts. If we let
1 and use the denition of /
0
and the initial condition c(0) = c
0
, we
get (IBC) if and only if (NPG) holds. It follows that when (NPG) is satised
with strict equality, so is (IBC), and vice versa.
An alternative proof is obtained by using the general solution to a linear
inhomogeneous rst-order dierential equation and then let 1 . Since
this is a more generally applicable approach, we will show how it works and
use it for Claim 1 below (an extended version of Proposition 1) and for the
proof of Proposition 2. Claim 1 will for example prove useful in Exercise 9.1
and in the next chapter.
CLAIM 1 Let ,(t) and q(t) be given continuous functions of time, t. Consider
the dierential equation
r(t) = q(t)r(t) + ,(t). (9.59)
with r(t
0
) = r
t
0
. a given initial value. Then the inequality
lim
t
r(t)c
I
I
0
j(c)oc
0 (9.60)
364
is equivalent to
Z

t
0
,(t)c
:
I
0
j(c)oc
dt r
t
0
. (9.61)
Moreover, if and only if (9.60) is satised with strict equality, then (9.61) is
also satised with strict equality.
Proof. The linear dierential equation, (9.59), has the solution
r(t) = r(t
0
)c
I
I
0
j(c)oc
+
Z
t
t
0
,(t)c
I
:
j(c)oc
dt. (9.62)
Multiplying through by c
I
I
0
j(c)oc
yields
r(t)c
I
I
0
j(c)oc
= r(t
0
) +
Z
t
t
0
,(t)c
:
I
0
j(c)oc
dt.
By letting t . it can be seen that if and only if (9.60) is true, we have
r(t
0
) +
Z

t
0
,(t)c
:
I
0
j(c)oc
dt 0.
Since r(t
0
) = r
t
0
. this is the same as (9.61). We also see that if and only if
(9.60) holds with strict equality, then (9.61) also holds with strict equality.
COROLLARY Let : be a given constant and let

/
t
0

Z

t
0
n(t)c
:
I
0
(v(c)a)oc
dt. (9.63)
which we assume is a nite number. Then, given
c(t) = (:(t) :)c(t) + n(t) c(t). where c(t
0
) = c
t
0
, (9.64)
it holds that
lim
t
c(t)c
I
I
0
(v(c)a)oc
0
Z

t
0
c(t)c
:
I
0
(v(c)a)oc
dt c
t
0
+ /
t
0
. (9.65)
where a strict equality on the left-hand side of implies a strict equality
on the right-hand side, and vice versa.
Proof. Let r(t) = c(t). q(t) = :(t) : and ,(t) = n(t) c(t) in (9.59), (9.60)
and (9.61). Then the conclusion follows from Claim 1.
By setting t
0
= 0 in the corollary and replacing t by t and : by 0, we
have hereby provided an alternative proof of Proposition 1.
9.7. Appendix 365
D. Proof of Proposition 2 (the transversality condition with an
innite time horizon)
In the dierential equation (9.59) we let r(t) = `(t). q(t) = (:(t) j). and
,(t) = 0. This gives the linear dierential equation

`(t) = (j :(t))`(t).
which is identical to the rst-order condition (9.37) in Section 9.3. The
solution is
`(t) = `(t
0
)c
I
I
0
(v(c)j)oc
.
Substituting this into (TVC) in Section 9.3 yields
`(t
0
) lim
t
c(t)c
I
I
0
(v(c)a)oc
= 0. (9.66)
Fromthe rst-order condition (9.36) in Section 9.3 we have `(t
0
) = n
0
(c(t
0
))
0 so that `(t
0
) in (9.66) can be ignored. Thus (TVC) in Section 9.3 is equiva-
lent to the condition (NPG) in that section being satised with strict equality
(let t
0
= 0 = :).
E. Intertemporal consumption smoothing
We claimed in Section 9.4 that equation (9.45) gives approximately the mar-
ginal rate of substitution of consumption in the time interval (t+t, t+2t)
for consumption in (t, t + t). This can be seen in the following way. To
save notation we shall write our time-dependent variables as c
t
. :
t
, etc., even
though they are continuous functions of time. The contribution from the two
time intervals to the criterion function is
Z
t+2t
t
n(c
t
)c
jt
dt c
jt
Z
t+t
t
n(c
t
)c
j(tt)
dt +
Z
t+2t
t+t
n(c
t+t
)c
j(tt)
dt
= c
jt
n(c
t
)
c
j(tt)
j
t+t
t
+ n(c
t+t
)
c
j(tt)
j
t+2t
t+t
!
=
c
jt
(1 c
jt
)
j
n(c
t
) + n(c
t+t
)c
jt
.
Requiring unchanged utility integral l
0
=

l
0
is thus approximately the same
as requiring [n(c
t
) + n(c
t+t
)c
jt
] = 0. which by carrying through the
dierentiation and rearranging gives (9.45).
The instantaneous local optimality condition, equation (9.48), can be
interpreted on the basis of (9.47). Take logs on both sides of (9.47) to get
ln n
0
(c
t
) + jt ln n
0
(c
t+t
) =
Z
t+t
t
:
t
dt.
366
Dividing by t, substituting (9.46), and letting t 0 we get
j lim
t0
ln n
0
(c
t+t
) lnn
0
(c
t
)
t
= lim
t0
1
t+t
1
t
t
. (9.67)
where 1
t
is the antiderivative of :
t
. By the denition of a time derivative,
(9.67) can be written
j
d lnn
0
(c
t
)
dt
=
d1
t
dt
.
Carrying out the dierentiation, we get
j
1
n
0
(c
t
)
n
00
(c
t
) c
t
= :
t
.
which was to be shown.
F. Elasticity of intertemporal substitution in continuous time
The relationship between the elasticity of marginal utility and the concept
of instantaneous elasticity of intertemporal substitution in consumption can
be exposed in the following way: consider an indierence curve for consump-
tion in the non-overlapping time intervals (t, t + t) and (:, : + t). The
indierence curve is depicted in Fig. 9.3. The consumption path outside
the two time intervals is kept unchanged. At a given point (c
t
t. c
c
t) on
the indierence curve, the marginal rate of substitution of :-consumption for
t-consumption, `1o
ct
, is given by the absolute slope of the tangent to the
indierence curve at that point. In view of n
00
(c) < 0, `1o
ct
is rising along
the curve when c
t
decreases (and thereby c
c
increases).
Conversely, we can consider the ratio c
c
,c
t
as a function of `1o
ct
along
the given indierence curve. The elasticity of this consumption ratio w.r.t.
`1o
ct
as we move along the given indierence curve then indicates the
elasticity of substitution between consumption in the time interval (t, t +t)
and consumption in the time interval (:, : +t). Denoting this elasticity by
o(c
t
. c
c
). we thus have:
o(c
t
. c
c
) =
`1o
ct
c
c
,c
t
d(c
c
,c
t
)
d`1o
ct

(cscI)
c
s
c
I
A1S
sI
A1S
sI
.
At an optimum point, `1o
ct
equals the ratio of the discounted prices
of good t and good :. Thus, the elasticity of substitution can be interpreted
as approximately equal to the percentage increase in the ratio of the chosen
goods, c
c
,c
t
, generated by a one percentage increase in the inverse price
9.7. Appendix 367
Indifferencecurve
new c /
s t
c
c /
s t
c
Figure 9.3: Substitution of :-consumption for t-consumption as '1o
ct
increases
til '1o
0
ct
.
ratio, holding the utility level and the amount of other goods unchanged.
If : = t + t and the interest rate from date t to date : is :. then (with
continuous compounding) this price ratio is c
vt
. cf. (9.47). Inserting `1o
ct
from (9.45) with t +t replaced by :, we get
o(c
t
. c
c
) =
n
0
(c
t
),[c
j(ct)
n
0
(c
c
)]
c
c
,c
t
d(c
c
,c
t
)
d{n
0
(c
t
),[c
j(ct)
n
0
(c
c
)]}
=
n
0
(c
t
),n
0
(c
c
)
c
c
,c
t
d(c
c
,c
t
)
d(n
0
(c
t
),n
0
(c
c
))
. (9.68)
since the factor c
j(tc)
cancels out.
We now interpret the ds in (9.68) as dierentials (recall, the dierential of
a dierentiable function = ,(r) is denoted d and dened as d = ,
0
(r)dr
where dr is some arbitrary real number). Calculating the dierentials we get
o(c
t
. c
c
)
n
0
(c
t
),n
0
(c
c
)
c
c
,c
t
(c
t
dc
c
c
c
dc
t
),c
2
t
[n
0
(c
c
)n
00
(c
t
)dc
t
n
0
(c
t
)n
00
(c
c
)dc
c
],n
0
(c
c
)
2
.
Hence, for : t we get c
c
c
t
and
o(c
t
. c
c
)
c
t
(dc
c
dc
t
),c
2
t
n
0
(c
t
)n
00
(c
t
)(dc
t
dc
c
),n
0
(c
t
)
2
=
n
0
(c
t
)
c
t
n
00
(c
t
)
o(c
t
).
This limiting value is known as the instantaneous elasticity of intertemporal
substitution of consumption. It reects the opposite of the desire for con-
sumption smoothing. Indeed, we see that o(c
t
) = 1,o(c
t
), where o(c
t
) is the
elasticity of marginal utility at the consumption level c(t).
368
9.8 Exercises
9.1 We look at a household (or dynasty) with innite time horizon. The
households labor supply is inelastic and grows at the constant rate : 0.
The household has a constant rate of time preference j : and the individual
instantaneous utility function is n(c) = c
10
,(1 o). where o is a positive
constant. There is no uncertainty. The household maximizes the integral
of per capita utility discounted at the rate j :. . Set up the households
optimization problem. Show that the optimal consumption plan satises
c(0) = ,
0
(c
0
+ /
0
). where
,
0
=
1
R
0
c
I
0
(
(10)r(:)p
0
+a)ot
dt
. and
/
0
=
Z

0
n(t)c
I
0
(v(t)a)ot
dt.
where n(t) is the real wage per unit of labor and otherwise the same notation
as in this chapter is used. Hint: use the corollary to Claim 1 in Appendix C
and the method of Example 1 in Section 9.5.

Solving The Intertemporal Consumption/saving Problem in Discrete and Continuous Time

Uploaded by

Copyright:

Available Formats

Solving The Intertemporal Consumption/saving Problem in Discrete and Continuous Time

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solving The Intertemporal Consumption/saving Problem in Discrete and Continuous Time

Uploaded by

Copyright:

Available Formats

Chapter 9

Solving the intertemporal

1(t) = 1(t) o1(t). 1(0) = 1

< 0, which violates (NPG).

`(t) = (j :(t))`(t). (9.37)

`,` by drawing on (9.36) again. Applying (9.37) now gives

EXAMPLE 2 (constant absolute semi-elasticity of marginal utility; innite

COROLLARY Let : be a given constant and let

You might also like