Output Dynamics with Heterogenous Agents:

Impact of Income/Wealth Inequality in Solow, R-C-K and OLG

Mausumi Das

Lecture 13, EC004, DSE

April 10, 2024

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 1 / 27
OLG Model: Basic Set Up (A Recap)

In the last class we discussed the basic set up of the baseline

(Samuelson-Diamond) OLG framework.
At time t, the economy starts with a given stock of capital (Kt ) -
provided by the current old, and a given stock of labour force (Nt ) -
provided by the current young.
These factors are supplied inelastically to competitive …rms who
generate the aggregate output using a neoclassical production
technology:
Yt = F (Kt , Nt )
At every point of time the market wage rate and the rental rate of
Kt
capital are given by wt = f (kt ) kt f 0 (kt ); rt = f 0 (kt ) where kt Nt .
Total output is distributed as wage income (going to the younger
generation) and interest income (going to the older generation):

Yt = wt Nt + rt Kt . (1)

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 2 / 27
OLG Model: Basic Set Up (Contd.)

The current old generation consumes not only the interest earnings
but the left over capital stock as well such that

ct2 1,t Nt 1 = rt Kt + (1 δ)Kt .

A young agent of the current generation on the other hand decides on

his present and future consumption so as to maximise his lifetime
utility:
1 2 1 2
U (ct,t , ct,t +1 ) u (ct,t ) + βu (ct,t +1 ); 0 < β < 1, (2)

subject to his life-time budget constraint:

2
ct,t
1 +1
ct,t + = wt . (3)
(1 + rte+1 δ)

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 3 / 27
OLG Model: Basic Set Up (Contd.)

Assuming that agents have perfect foresight/rational

expectations such that rte+1 = rt +1 for all t, the FOC of the
optimization exercise generates the following familiar Euler equation:

u 0 (ct,t
1 )

2
= β(1 + rt +1 δ ). (4)
u 0 (ct,t +1 )

From the FOC and the life-time budget constraint, we can derive the
optimal solutions as:
1
ct,t = ψ(wt , rt +1 );
2
ct,t +1 = η (wt , rt +1 ).

Corresponding optimal savings function:

st = wt ψ(wt , rt +1 ) φ(wt , rt +1 ).

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 4 / 27
OLG Model: Basic Set Up (Contd.)
In order to characterise the savings function of the young agents, we
analysed the signs of the partial derivatives sw and sr .
The signs of these partial derivatives would tell us how savings respond
to a change in the current wage rate (wt ) and a change in the future
interest rate (rt +1 ).
Under a generic utility function where both c 1 and c 2 are normal
goods, we found that
(1) The sign of sw is unambiguous:
0 < sw < 1.
(2) The sign of sr however is ambiguous: it depends on the relative
strength of the income e¤ect vis-a-vis the substitution e¤ect of a
price change. In particular,
sr R 0
according as, substitution e¤ect R income e¤ect.
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 5 / 27
OLG Model: Basic Set Up (Contd.)

Apriori we have no reason to rule out the case where sr < 0, even
though it is counter-intuitive.
However, henceforth we assume that the utility function of the
agents is such that the substitution e¤ect of a price change is at least
as strong as the corresponding income e¤ect, so that

sr = 0 (Assumption 1)

At this point, this is purely an assumption - without any

economic justi…cation. (We will provide a justi…cation soon.)

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 6 / 27
OLG model: Dynamics of Capital-Labour Ratio:
The path of capital accumulation for the economy was obtained by
aggregating the savings (st ) over the entire population of young
generation today (Nt ). Thus Kt +1 = st Nt .
Since labour force (i.e., young generation in every successive period)
is growing at the rate n, i.e., Nt +1 = (1 + n )Nt , the dynamic
equation for the capital-labour ratio (kt ) for this OLG economy was
obtained as:
Kt +1 st φ(wt , rt +1 )
kt +1 = = . (5)
Nt +1 (1 + n ) (1 + n )
Noting that wt = f (kt ) kt f 0 (kt ) and rt +1 = f 0 (kt +1 ), we rewrote
(5) as:
φ(wt (kt ), rt +1 (kt +1 ))
kt +1 = Φ(kt , kt +1 ). (6)
(1 + n )
Equation (6) is the basic dynamic equation of the OLG model, which
implicitly de…nes kt +1 as a function of kt . Given k0 , we should be able
to trace the evolution of the capital-labour ratio over time.
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 7 / 27
Existence of a Unique Perfect Foresight Path:

The dynamic equation (6) is an ‘implicit’di¤erence equation with

kt +1 entering on both sides.

In fact this implicit nature of the equation arises precisely due to

assumption of perfect foresight.

This then raises the following question: given a kt , do we necessarily

get a unique kt +1 > 0 that satisfy the dynamic equation?

In other words, given any kt , does a unique perfect foresight path

exist such that the dynamic equation (6) de…nes a unique
non-negative kt +1 for all subsequent time periods?

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 8 / 27
Existence of a Unique Perfect Foresight Path: (Contd.)

Notice that existence and uniqueness are both important -

otherwise the future trajectory of the economy will become
indeterminate.

Indeed, for any given value of kt , unless we can specify a unique

non-negative value of kt +1 for the next period, the di¤rence equation
(6) is not even well-de…ned!!

And if the di¤erence equation itself is not valid, the question of

identifying its time path (solution) and its long run properties (such
as, steady state and stability) does not even arise!!!

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 9 / 27
Existence of a Unique Perfect Foresight Path: (Contd.)
Mathematically, the problem boils down to the following question:
Consider a given value of kt = k̄ > 0.
Does the following equation generates a unique non-negative solution
for kt +1 ?
φ(wt (k̄ ), rt +1 (kt +1 ))
kt + 1 = Φ(k̄, kt +1 ) (7)
(1 + n )
Given k̄, both the LHS and RHS of (7) become functions of kt +1
alone. For a unique solution, we need a unique intersection point
between the LHS and the RHS.
Obviously, in order to ensure that we have to impose conditions on
the Φ(k̄, kt +1 ) function so that it is "well-behaved".
Note that the LHS of (7) maps kt +1 unto itself and therefore will be
represented by a 45o line.
The RHS is a related to kt +1 through the young agent’s savings
function: st = φ(wt , rt +1 ). Hence it’s slope and curvature will depend
on how the savings function responds to changes in kt +1 .
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 10 / 27
Existence of a Unique Perfect Foresight Path (Contd.):
Indeed, given kt = k̄, from (7) we shall have a unique non-negative
solution for kt +1 if and only if the curve representing Φ(k̄, kt +1 ) has a
unique point of intersection with the 45o line in the positive quadrant.
A su¢ cient condition for this to happen is Φ(k̄, kt +1 ) is either a ‡at
line or is downward sloping with respect to kt +1 , i.e.,
∂Φ(k̄, kt +1 )
5 0 for any given k̄.
∂kt +1
This case is depicted in the diagram below:

(Please note that this is NOT the phase diagram!!!)

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 11 / 27
Existence of a Unique Perfect Foresight Path (Contd.):
Notice that
∂Φ(k̄, kt +1 ) 1 ∂φ(wt , rt +1 )
=
∂kt +1 (1 + n ) ∂kt +1
1 ∂φ(wt , rt +1 ) drt +1
=
(1 + n ) ∂rt +1 dkt +1
1
= sr f 00 (kt +1 ).
(1 + n )
Thus a su¢ cient condition for the existence of a unique perfect
foresight path is: sr = 0.
Indeed this is ensured in our model by Assumption 1.
Thus Assumption 1 (speci…ed earlier without any explicit theoretical
rationale) ensures that a unique perfect forsight path exists and the
dynamic equation given by (6) is a valid dynamic equation that
characterises the evolution of the capital-labour ratio for this OLG
economy starting from any initial k0 .
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 12 / 27
Unique Perfect Foresight Path: OLG vs R-C-K
Notice uniqueness of the perfect foresight path is an issue in the OLG
model, because there is a circularity involved: savings (st ) depends on
kt +1 (through anticipated return rt +1 ); again kt +1 depends on st
(actual capital formation).
This mutual interdependence between st and kt generates the
possibility of multiple solution paths - all consistent with rational
self-ful…lling expectations.
So we need additional assumptions (like sr = 0) to rule this out!
Indeed, sr = 0 may not be satis…ed with even utility functions
belonging to CRRA variety! So the OLG model requires more
stringent conditions on the utility function than the R-C-K model.
For example, consider the following CRRA utility function:
1 )1 2
(ct,t 1
+1 )
θ
1 2 (ct,t θ
U (ct,t , ct,t +1 ) +β ; θ 6= 1.
1 θ 1 θ
You can show that in this case sr < 0 if θ > 1 (Homework).
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 13 / 27
Unique Perfect Foresight Path: OLG vs R-C-K (Contd.)
But we had a similar inter-dependence in the R-C-K model too!
Recall that the dynamic equations for the economy in the R-C-K V1
was given by:
2 3
6 7
ct +1 = β 41 + α (kt +1 )α 1 δ 5 ct ;
| {z }
r t +1
α
kt +1 = (kt ) + (1 δ) kt ct .
The Euler equation determines a houshold’s current consumption-
savings choice, based on anticipated return rt +1 ; again the actual
return tomorrow is determined by kt +1 - which in turn depends on the
household’s consumption-savings choice - as captured by the level of
ct .
So why doesn’t it result in multiple possible rational expectactions
paths in the R-C-K model?
Or does it???
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 14 / 27
Unique Perfect Foresight Path: OLG vs R-C-K (Contd.)
Recall the phase diagram of the R-C-K model:

Recall that there were indeed multiple dynamic paths - all consistent
with the two dynamic equations which are derived based of rational
expectation/perfect foresight!
However we could rule out all but one - by applying the ini…nite
horizon transversality condition!
In the OLG model, there is no such ‘ini…nite horizon’transversality
condition! Indeed there is nothing to bind their expectations - and
therefore their savings behaviour - to a speci…c "well-behaved" long
run path!
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 15 / 27
OLG Model: Dynamics of Capital-Labour Ratio Revisited

Now back to the OLG model.

Having now established that given our assumption that sr = 0, the
dynamic equation given by (6) is a valid dynamic equation under
perfect foresight, let us now characterise the evolution of kt over time
starting from any initial k0 .
The questions we now address are as follows:
(1) Does the dynamics of capital-labour ration in this OLG economy admit
a non-trivial steady state?
(2) If yes, is that non-trivial staedy state unique?
(3) Is that steady state stable?
(4) Is that steady state necessarily dynamically e¢ cient?

Let us now answer these questions one by one.

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 16 / 27
OLG Model: Dynamics of Capital-Labour Ratio (Contd.)
Recall the dynamic equation of the OLG model:
φ(wt (kt ), rt +1 (kt +1 ))
kt +1 = Φ(kt , kt +1 ).
(1 + n )
Since Φ(kt , kt +1 ) is a nonlinear function of kt and kt +1 , we shall
have to use the phase diagram technique to qualitatively characterise
the dynamics.
In drawing the phase diagram, …rst note that the slope of the phase
line can be determined by total di¤entiating both sides:
dwt drt +1
(1 + n)dkt +1 = sw dkt + sr dkt +1
dkt dkt +1
dkt +1 sw [ kt f 00 (kt )]
i.e., = .
dkt (1 + n) sr f 00 (kt +1 )
Under the assumption that sr = 0, the slope of the phase line is
necessarily positive. But what about it’s curvature?
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 17 / 27
Dynamics of Capital-Labour Ratio (Contd.):
Even when the slope of the phase line is positive, the curvature is not
necessarily concave - since it would involve the third derivative of the
utility function and the f (k ) function - whose signs are not known.
Hence anything is possible: we may have situations of no steady
state; unique stable steady state; unique unstable steady state;
multiple steady states (some stable, some unstable):

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 18 / 27
Uniqueness and Stability of Steady State in the OLG
Model:

In other words, the nice result of the Solow & R-C-K model of a
unique and stable steady state is no longer gauranteed - despite the
production function satifying all the standard neoclassical
properties - including diminishing returns and the Inada
conditions!
Indeed the OLG model opens up possibilities of non-convegence and
poverty traps:
Two economies starting with di¤erent initial capital-labour ratios may
end up at two completely di¤erent steady state levels of per capita
income in the long run- despite having identical parameters.

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 19 / 27
Dynamic E¢ ciency in the OLG Model:

What about dynamic e¢ eciency?

Even that is no longer gauranteed!
We will provide an example below to show that dynamic e¢ ciency is
not necessarily gauranteed under the OLG model - despite optimizing
savings behaviour by the agents.
But before we turn to the speci…c counter example, we need to
characterise dynamic e¢ eciency in the context of the OLG model.

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 20 / 27
Golden Rule & Dynamic E¢ ciency in the OLG Model:

Recall the we had earlier de…ned dynamic e¢ ciency in the context of

the golden rule condition, given by

kg : f 0 (k ) = (n + δ)

Indeed, any steady state k > kg was identi…ed as dynamically

ine¢ cient!
Also recall that the ‘golden rule’is de…ned as that particular steady
state which maximises the steady state level of per capita
consumption.
In the Solow model, the steady state level of per capita consumption
is: c = (1 s )f (k ), which is identical for everybody.
But in the OLG model, we have two di¤erent sets of people: the
young and the old.
So whose steady state consumption level are we talking about?
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 21 / 27
Golden Rule & Dynamic E¢ ciency in the OLG Model:
(Contd.)

Indeed, what would be the corresponding golden rule value in the

OLG model that maximises steady state "per capita" consumption?
Note that in an OLG model, per capita (average) consumption at any
point of time t would be de…ned as:
1 N + c2
ct,t 1 N + c2 1 + c2
(1 + n)ct,t
t t 1,t Nt 1 ct,t t t 1,t Nt 1 t 1,t
c̄t = = =
Lt Nt + Nt 1 (2 + n )

where Nt = (1 + n )Nt 1.
Indeed the per capita consumption is an weighted average of the
consumption level of the young and the consumption level of the old.
Given this de…nition, what would be the steady state value of per
capita (average) consumption?

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 22 / 27
Golden Rule & Dynamic E¢ ciency in the OLG Model:
(Contd.)
The basic dynamic equation is the OLG model is given by:
st φ(w (kt ), r (kt +1 ))
kt +1 = =
(1 + n ) (1 + n )
Accordingly, the steady state(s) of the OLG model is de…ned as:
s φ(w (k ), r (k ))
k = =
(1 + n ) (1 + n )
) s = (1 + n )k (8)
Hence, from the optimality conditions of c 1 and c 2 , we know that at
steady state:
c1 = ψ(w (k ), r (k )) = w (k ) s ; (9)
2
c = η (w (k ), r (k )) = [1 + r (k ) δ] s (10)

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 23 / 27
Golden Rule & Dynamic E¢ ciency in the OLG Model:
(Contd.)
Hence from (8), (9) & (10), the steady state per capita consumption
in the OLG model is given by:
(1 + n ) c 1 + c 2
c =
(2 + n )
(1 + n) [w (k ) s ] + [(1 + r (k ) δ)] s
=
(2 + n )
(1 + n)w (k ) + [r (k ) δ n)] s
=
(2 + n )
(1 + n) [fw (k ) + r (k )k g (n + δ)k ]
=
1 + (1 + n )
1+n
= [f (k ) (n + δ )k ]
(2 + n )
Maximizing c with respect to k , we would get the same golden rule
condition as before: kg : f 0 (k ) = (n + δ).
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 24 / 27
Golden Rule & Dynamic E¢ ciency in the OLG Model:
(Contd.)

Indeed, any steady state k > kg is again identi…ed as dynamically

ine¢ cient!

By moving from k to kg , you will have higher average consumption

level, which you can allocate between the young and old making both
happier (pareto improvemnet)!

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 25 / 27
Dynamic Ine¢ ciency in the OLG Model: Example with
Speci…c Functional Form
We now provide a counter-example to show that even dynamic
e¢ eciency is no longer guranteed in the OLG model.
For this purpose, we consider some speci…c functional forms. Let
1 2
U (ct,t , ct,t +1 ) log ct1 + β log ct2+1 ;
f (kt ) = (kt )α ; 0 < α < 1.
Also let the rate of depreciation be 100%, i.e., δ = 1.
Homework for you:
(i) Solve for the optimal consumption and savings of the young
generation for this speci…c utility function.
(ii) Derive the correspnding dyamic equation for kt and verify that the
su¢ cient condition for existence of unique perfect foresight path
exists.
(iii) Derive the steady non-trivial state and verify that it is unique and
stable. Show that it may still not be dynamically e¢ cient!
Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 26 / 27
References:

Reference for the Baseline OLG Model:

D. Acemoglu: Introduction to Modern Economic Growth,
Chapter 9, Sections 9.2,9.3, 9.4.
O.Galor & H. Ryder (1989): "Existence, uniqueness, and
stability of equilibrium in an overlapping-generations model with
productive capital,", Journal of Economic Theory, pp.360–375.

Das (Lecture 13, EC004, DSE) Heterogenous Agents: Solow to R-C-K to OLG April 10, 2024 27 / 27