The Geneva Papers on Risk and Insurance Theory, 26: 5–24, 2001


c 2001 The Geneva Association

Social Security with Heterogeneous Populations

Subject to Demographic Shocks
GABRIELLE DEMANGE demange@[Link]
DELTA, 48 Boulevard Jourdan, Paris 75014, France

GUY LAROQUE laroque@[Link]

INSEE, Direction des Etudes et Synthèses Economiques, 15 boulevard Gabriel Péri,
BP 100, 92244 Malakoff Cedex, France

Abstract
In a previous paper, we showed how a pay-as-you-go social security scheme, based on voluntary contributions,
can be an appropriate institution to reach an optimal sharing of risks among generations in the presence of
demographic uncertainties. We study here the functioning of such schemes when there are different population
strata, with different demographic shocks and wages. We show that while a collective voluntary pay-as-you-
go scheme can provide efficient intergenerational risk sharing, it is likely to be destabilized by pensions funds
specialized by agents’ types. This is true both when there is a complete set of contingent markets, where the risk
pooling capabilities of a collective fund are potentially of less interest, and when markets are incomplete. In this
last circumstance, a collective fund may help the living agents to share their intragenerational risks. However, we
show that the resulting allocation does not Pareto dominate the outcome of individual funds by agent types, and
that there are incentives for agents to separate from any collective organization.

Key words: intergenerational risk sharing, pay-as-you-go redistribution schemes

JEL classification: E2, E6, H3

1. Introduction

The social security system, which implements a redistribution of wealth across generations,
is expected to face severe strain. The baby boom generations believe that the return on their
contributions will be unsatisfactory, at least in comparison with investments on the stock
markets, since the next generations to come are relatively less numerous. These difficulties
are particularly severe for some defined benefits funds, designed to accommodate the speci-
ficities of various professions whose numbers have declined, and which find themselves
unable to serve the promised pensions. The rationale for the existence of multiple funds
therefore has been put to question. Indeed, a socially organized collective system seems
to automatically provide insurance to the members of a generation against the microeco-
nomic idiosyncratic shocks that they may face. The aim of this paper is to make progress
in the understanding of the optimality properties of a social security system possibly based
on different funds. This is a question of some practical importance, witness the Chilean
6 GABRIELLE DEMANGE AND GUY LAROQUE

reform a major ingredient of which was to create competition between pension funds [see
for instance Edwards, 1996].
In a previous paper [Demange-Laroque, 2000], we show how a pay-as-you-go social
security scheme, based on voluntary contributions, can be an appropriate institution to reach
an optimal sharing of risks among generations in the presence of demographic uncertainties.
These schemes, in contrast with commonly used systems, are not defined benefit plans: the
pensions received depend on the voluntary contributions of the active population, which
in turn depend on the current incomes and on the (rationally) anticipated future pensions.
We study here the functioning of such schemes when there are different population strata,
associated with different demographic and wage shocks.
To address these questions we consider an overlapping generations model with different
types of agents and a single physical good. There are macro economic shocks on the return
to capital, on wages, and on population growth. The shocks are Markov, and such that
the distribution of population by types is stationary. We limit our attention to stationary
allocations of resources. We leave for further research the study of the transition between
stationary allocations.
To compare the various possible social security institutions, we use the concept of interim
optimality introduced by Peled1 [1984], which amounts to standard Pareto optimality once
the state of nature in which the agents are born is known. There are three conditions for the
optimality of a stationary allocation. First, the level of investment in physical capital must
be optimal, in the sense that the expected return on a unit of capital must equal the marginal
cost supported today. Second, the living generations must optimally share the risks that they
bear. Finally, some transfers across generations are in order, to allocate the demographic
risks.
We show that a collective voluntary pay-as-you-go scheme, as in the case of an homoge-
neous population, can provide efficient intergenerational risk sharing, if it is supplemented
with a complete set of markets organized between the living agents at each date. More gen-
erally, with such a complete set of markets, any viable system of voluntary pay-as-you-go
schemes, provided that each category of the population is affiliated with at least one such
fund, would also yield the same allocation. Therefore there is no ground to favor a collective
scheme over specific funds.
This result extends, somewhat surprisingly, to situations where the asset markets are
incomplete. Let the same (incomplete) set of financial assets be given. Consider an equilib-
rium allocation associated with a collective social security system, and one associated with
population specific funds. The equilibrium allocations are typically not interim optimal any
longer. But again, the two allocations are not comparable: going from one to the other, there
is always, for each population strata, a state of nature under which it loses some expected
welfare. Since, at any time, there are generically some groups of agents that would benefit
from leaving a collective arrangement, the institution of population specific funds seems to
be more stable than a collective fund.
We assume that the voluntary contributions to the funds are positive at the equilibrium.
When there is an homogeneous population and no uncertainty (Samuelson model), a col-
lective fund with positive voluntary contributions is another interpretation of a monetary
equilibrium. It is well known [Gale, 1973] that a necessary and sufficient condition for
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 7

existence of an equilibrium with positive transfers from the young to the old generation is
that the return on capital be smaller than the population growth rate at the autarky (i.e. without
intergenerational transfers) equilibrium. We have extended this result to our stochastic setup
with an homogeneous population (but not for heterogeneous populations yet), where the
condition can be stated in terms of a matrix made of the marginal rates of substitution
between the future and current states, weighted by the rate of growth of population [see
also Manuelli, 1990 for a similar result but without production and demographic shocks].
The model is laid down in the next section. Then, we discuss the optimality properties of
stationary allocations. In Section 4, we describe the equilibria associated with a collective
pay as you go system, and study its optimality properties. Section 5 considers multiple
funds. The final section contains the proofs.

2. The model

To study the sharing of risks between generations, we take a simple model, where the over-
lapping generations live for two periods, earn and consume a single good. The exogenous
shocks affect the rates of growth of the populations, (labor) incomes, and the rates of return
on savings. The shocks are assumed to follow a first order Markov process: the distribution
of shocks at date t + 1 conditional on what happened at date t is invariant over time.
The demographics There are several types of agents, indexed by i, i = 1, . . . , I . Each agent
i is representative of a lineage of agents who share the same characteristics, preferences
and wages. The evolution of the distribution of agents by type depends on the (random)
growth factors that affect each category. Let µi− be the proportion of category i in the
population at some date t − 1 and γ i the growth factor of population i between t − 1 and
t. The growth of the aggregate population and its composition at date t then are given
by:2

I
γi
γ = µi− γ i and µi = µi− . (1)
i=1
γ

The agents characteristics The representative agent of type i earns an income wi from his
labor when young. The utility he gets from his lifetime consumption profile (a i , bi ) is
denoted u i (a i , bi ). It is defined on R2+ , is concave, strictly increasing in its two arguments
and twice continuously differentiable on R2++ . To avoid corner solutions, we assume
 
lima→0 u ia = ∞ and limb→0 u ib = ∞.
The technology The good may be invested in a linear technology with a random rate of
return ρ, identical for all types. An investment s i made by a young agent yields ρ+ s i in
the next period if ρ+ is realized.
The feasibility constraints In category i, there are γ i young agents for each old one. There-
fore the feasibility constraint at date t, per head of old agent, is:


I
    I
 i 
µi− γ i ati + sti + bti = µi− ρst−1 + γ i wi (2)
i=1 i=1
8 GABRIELLE DEMANGE AND GUY LAROQUE

The left hand side sums up the current consumptions and investments. It has to be equal
to the resources coming from previous investments and current wages, the right hand
side.
The stochastic structure It is convenient to describe the exogenous characteristics of the
economy at the current date as the ‘state of the economy’. Formally, this is a vector x =
(µi , γ i , wi , i = 1, . . . , I , ρ), made of the proportions of young agents in the population,
the growth factors,3 wages and the return on investment. We assume that, at the time of
the consumption and saving decisions, the information set of the agents is equal to the
current state.
The state is assumed to follow a first order Markovian process, with an exogenous
transition matrix Pr and a unique invariant distribution. The support of the distribution,
denoted by X , only contains strictly positive vectors. For simplicity, we assume that the
state space X is finite.4 As far as rates of return and wages are concerned, this assumption
is innocuous. For the dynamics of the population, it implies that, although the size of
the aggregate population may become indefinitely large or shrink to zero, the relative
proportions of the various types of agents stay bounded.
We shall denote a coordinate of x by its value, say µi or more explicitly µi (x) if
necessary. Similarly for aggregates, say µ or µ(x) and previous or future variables, µi−
or µi (x− ), µi+ or µi (x+ ), where x− (resp. x+ ) is the predecessor (resp. successor) of
state x.
Stationary allocations The young agents’ decisions, their consumption a i and investment
s i , are functions of their information. Given the Markovian structure it is natural to
consider that they are stationary functions of the state. Given stationary decisions rules
a i , s, the total available
 I resources in state x per head of the currently living old generation
are then equal to i=1 µi− [ρs i (x− ) + γ i wi ]. They depend on the previous state x−
I
through the average previous investment s(x− ) = i=1 µi− s i (x− ). Therefore feasibility
of a stationary allocation requires that the old’s consumption in state x depends on the
previous state x− : it is a function bi (x− , x). Using (1), i.e. µi− γ i = µi γ , the feasibility
constraint (2) may be written as:


I 
I
γ µi [a i (x) + s(x) − wi ] + µi− bi (x− , x) − ρs(x− ) = 0, (3)
i=1 i=1

for all (x− , x) where x is a successor of x− .

3. Optimality

3.1. Intergenerational transfers

Without uncertainty, it is well known that intergenerational transfers may be welfare improv-
ing. Here the situation is complicated by risk sharing considerations. Agents face the risks
associated with the return on their savings5 as well as, whenever some intergenerational
redistribution scheme is implemented, the risks coming from both the composition and
incomes of the next generation.
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 9

It is useful to give an explicit definition of the intergenerational transfers. Consider a

stationary allocation and define the average transfer (positive or negative) paid by the
young agents in state x as


I
τ (x) = µi [wi − a i (x) − s(x)]. (4)
i=1

Feasibility (3) then can be rewritten as


I
µi− bi (x− , x) = ρ(x)s(x− ) + γ (x)τ (x) (5)
i=1

which says that the average transfer to the old is γ (x)τ (x) if state x occurs.
The intergenerational transfers associated with a feasible allocation are naturally de-
scribed in terms of a social security institution. The agents contribute to the social security
fund when young, and receive pensions when old. The feasibility constraints (4) and (5)
can be interpreted as if, in the aggregate, s(x) + τ (x) is the contribution of the generation
born in state x to the system, with s(x) the funded part and τ (x) the unfunded part, and
ρ(x+ )s(x) + γ (x+ )τ (x+ ) is the benefit that this generation gets when old if x+ is realized.
In practice, these transfers typically vary across the various strata of the population and are
associated with a variety of incentives or regulatory constraints. In this paper we are mainly
interested in studying optimality of risk sharing and equilibria associated with different
social security systems, i.e. insurance between generations and groups.
Example. A simple example helps to understand some basic trade offs, and will be useful to
illustrate the results. Consider a situation where there is no capital, no aggregate uncertainty:
the population size, its composition, and  the aggregate wage are constant, but categories
suffer from idiosyncratic shocks, i.e. w = i wi (x) is constant independent of x but wi (x)
is random and varies across categories.
The agents have identical Cobb Douglas utility functions a 1−β bβ . The only income when
old comes from their contributions to a voluntary pay as you go social security system.
Benefits are proportional to contributions, the total receipts of the system at any date being
distributed to the current living olds. Then the young optimal consumption is:6

a i (x) = (1 − β)wi (x),

independently of the specific definition of the benefits. Consider two polar organizations:
either a collective fund gathers all agents or there are multiple funds, one for each category.
The total receipts of a pension fund are distributed to the subscribers proportionately to
their initial contributions. Next period pensions are respectively equal to:7
 i
βw (x)

b (x, x+ ) = or
i

 i
βw (x+ ).
10 GABRIELLE DEMANGE AND GUY LAROQUE

Since the individual consumption of the young agents and the aggregate pensions are
identical in both regimes, the allocations only differ in the distribution of these pensions.
Are they optimally distributed? Are the two systems comparable?
A trivial remark is that, given the state x, the idiosyncratic risk associated with the future
wages is not diversified away through the specialized funds. According to the mutuality
principle, all categories could be made better off: simply give category i the expected value
of its pension, namely βE[wi (x+ ) | x], which is feasible in the absence of macroeconomic
risk. On the contrary, the collective system provides optimal insurance of the idiosyncratic
risks. One might think that the return on the collective fund, which is equal to 1 with certainty,
is to be preferred to the specialized fund, which is random, equal to wi (x+ )/wi (x), with
unconditional mean approximately equal to 1 (to keep stationarity).
But there always exists some state at which category i has an incentive to separate from
the others and would rather contribute to a specialized fund. Indeed let consider a state x
for which wi (x) is minimal; then agent i expects his followers of the same type to be surely
richer than himself: the distribution of βwi (x+ ) knowing x dominates βwi (x). Therefore,
from the point of view of the living agents, who know the state x in which there are born, the
collective system is not unanimously preferred to specialized funds. We shall study whether
this property holds in general (see Theorem 6 below).
One may want to compare the social security systems from an ex ante point of view. The
ex ante utility levels, up to a multiplicative constant,8 are equal to E[wi (x)] for a collective
system. As for the specialized fund, they are given by E{wi (x)1−β E[wi (x+ )β | x]}, which
can be rewritten as
 β  β
E[wi (x)1−β ]E wi (x+ ) + cov wi (x)1−β , wi (x+ ) .

β β
Because of stationarity E[wi (x+ ) ] = E[wi (x) ] so that by Jensen inequality the first
term is surely smaller than E[wi (x)]. Therefore if the covariance term is negative, i.e.
successive wages are negatively correlated, the collective system is surely preferred ex
ante to the specialized one. This would plead, as conform with the basic intuition, in
favor of the collective system. However, the empirical evidence points to positive auto-
correlation of individual wages. Moreover, ex ante optimality is too strong a requirement.
Indeed the collective system does not lead to an ex ante optimal allocation: distributing
((1 − β)Ewi , βEwi ) is preferred on an ex ante basis to the collective system. But such
an allocation is not implementable whenever the young agents know their wages at the
time of their saving decisions. More generally, from the point of view of the living agents,
the current state x is given, and the decisions are based on the conditional distribution of
the return. The relevant criterion is therefore interim optimality, that is optimality con-
ditional on the state of the economy rather than the overly strong criterion of ex ante
optimality.

3.2. Interim optimality

Interim optimality requires that it be not possible to increase the expected utility of the
agents born in a given state of nature without lowering that of the agents born in some other
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 11

state. The expected utility is computed using the transition probability of the process, and
is equal to E{u i [a i (x), bi (x, x+ )] | x} for an agent born in state x. Formally:

Definition 1: A feasible stationary allocation is interim optimal if it maximizes:


λi (x)E{u i [a i (x), bi (x, x+ )] | x},
i,x

over the set of feasible allocations, for some strictly positive9 weights λi (x).

3.3. Boundedness

Due to the constant returns to scale technology, without further assumptions, the set of feasi-
ble consumption allocations may be unbounded above. For example assume no uncertainty
on returns nor on growth. If the return on capital is strictly larger than the growth rate of
population, there does not exist a stationary optimal allocation with a finite stock of capital:
any feasible stationary allocation with a finite stock of capital is dominated by another
stationary allocation with a larger stock of capital. One may show that in the presence of
uncertainty, a necessary and sufficient condition for bounded payoffs is that the returns on
capital do not ‘dominate’ the population growth rates, in the following sense.
Let us define a cycle as a sequence of states (x1 , x2 , . . . , xn+1 = x1 ), for some n ≥ 1,
such that each state can be reached with positive probability from the preceding one:10

Pr(xi+1 | xi ) > 0 for all i = 1, . . . , n.

The set of feasible expected payoffs associated with stationary allocations is bounded if
and only if there is at least a cycle (x1 , x2 , . . . , xn+1 ) such that

n 
n
γ (xi ) > ρ(xi ).
i=1 i=1

Remark that this condition does not preclude that in some state it is known for sure that
the next period growth rate is dominated by the returns on capital i.e. that

γ (x+ ) < ρ(x+ ) for all x+ such that Pr(x+ | x) > 0.

In the rest of the paper, we assume that the set of stationary feasible allocations is bounded.

3.4. Characterization of optimality

The interim optimality program amounts to maximizing a concave function on a closed

convex set, and is characterized by first order conditions. To simplify notations denote

u bi (a i (x), bi (x, x+ )) Pr(x+ | x)
mrsi (x, x+ ) = 
E u a i (a i (x), bi (x, x+ )) | x
12 GABRIELLE DEMANGE AND GUY LAROQUE

where u ia (a i (x), bi (x, x+ )) denotes the partial derivative with respect to the first argument
and similarly for b. This expression is the marginal rate of substitution for a young agent
born in state x between current consumption and consumption next period contingent on
the future state x+ .

Theorem 1: An allocation is interim optimal if and only if it satisfies the three following
conditions:

• x+ ρ(x + ) mrs (x, x + ) ≤ 1, with equality if s (x) > 0,
i i
(6)
• mrs (x, x+ ) is independent of i for all (x, x+ ),
i
(7)
• Define the positive matrix A by A(x, x+ ) = γ (x+ ) mrsi (x, x+ ), which is independent of
i by (7).
The largest eigenvalue of A is equal to 1. (8)

These conditions are associated respectively with the optimality of investment, intragen-
erational and intergenerational consumption sharing. Condition (6) is standard. Condi-
tion (7) asserts that the generation born in state x optimally allocates its aggregate resources
for consumption over both periods of its life. The optimality of investment decision, given
by (6), already implies a relationship between the marginal rates of substitution across
agents. But it asks for more: the risks associated with future consumption are optimally
shared among the young living agents, i.e. the mrsi (x, x+ ) are also equalized across agents
(similar conditions were first derived by Borch). Note that these conditions only bear on a
given generation, and do not put any restriction on intergenerational transfers.
The optimality of such transfers is associated with condition (8). If it is not met but (7)
holds, there is an intergenerational transfer which increases the utilities of all the young
agents whatever the state. Indeed consider a marginal transfer of consumption between the
young and old generations, identical for all agents in a given state: da i (x) = −dτ (x). Keep
the investment fixed, and change the consumption of the old by γ (x)dτ (x) so as to satisfy
feasibility. If the young agents correctly anticipate the induced changes in their old age,
they all agree to this modification if and only if

−dτ (x) + γ (x+ ) mrsi [x, x+ ] dτ (x+ ) ≥ 0, all x (9)
x+

or equivalently, when the marginal rates of substitution are equal, if

A dτ ≥ dτ. (10)

Since the matrix A has all its elements positive its maximal eigenvalue, say α, is positive and
associated with a positive eigenvector, say (τ (x)). If α is larger than 1, then (10) is met
for dτ equal to τ , which implies that the young’s consumption is decreased in every state
for dτ positive small enough proportional to τ . An increase in unfunded social security
is beneficial to all the young agents. The argument can be reversed if α is smaller than 1.
However there is a noticeable difference between the two cases. If α is larger than 1, the
new consumption sharing rule is not only beneficial to the young whatever the state, but
also to the currently living old because it increases the old’s consumption in the transition
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 13

to the new regime. If the eigenvalue is smaller than 1, the old generation objects to the new
sharing rule.
In the standard overlapping generations model without uncertainty, restricting to constant
allocations, the two cases respectively correspond to the Samuelson and classical cases,
whereas the optimality condition (8) yields the golden rule.11 Theorem 1 extends this
condition to the study of optimality of stationary allocations in a model with uncertainty
[similar results have been obtained by Peled [1984] with a slightly different criterion which
takes into account the first generation], with the additional requirement of optimal risk
sharing between the young agents.
In the overlapping generations models without uncertainty, an equilibrium is not always
optimal. The basic reason is the absence of appropriate transfers between generations.
An infinitely lived financial asset, money or government debt, allows such transfers to be
implemented and may restore optimality. We study whether equilibria are optimal when
there are risks bearing on population growth, income and production, building on the existing
literature. With a representative agent at each date, rational expectations equilibria in which
a financial asset is traded yield interim optimal allocations [see Aiyagari–Peled, 1991 and
Demange–Laroque, 1999, respectively without and with shocks on population growth]. In
a previous paper Demange–Laroque [2000], we showed that a voluntary pay-as-you-go
pension fund can stand for the financial asset, when it is in positive quantity. This seems
to be a particularly suitable institutional setup in the presence of demographic shocks. Our
aim here is to study the situation where the population is heterogeneous.
The analysis a priori depends upon the optimality of risk sharing between the young
agents, i.e. on the equality of the marginal rates of substitution, as given by (7). In situations
where intragenerational optimality holds because there are enough contingent markets, a
simple pay as you go system, as will be shown in the next section, may allow to decentralize
an optimal allocation. In other more realistic cases, the pay as you go system interferes with
the insurance needs of the agents. A second best analysis will then be conducted.

4. Voluntary collective PAYG social security

We shall consider pay-as-you-go systems that are all voluntary. They may be either collective
as in this section, or separate between the different groups, as in Section 5. The young agents
can contribute a sum of their choice to some social security funds, which guarantees them
when old a fraction, proportional to their initial subscriptions, of the contributions of the
next generation.
We first explain the simple case of a homogeneous population. If τ (x) is the subscription
of the generation born in state x, the return r (x, x+ ) on a dollar subscribed in state x
when next period state is x+ is equal to γ (x+ )T (x+ )/T (x). In an equilibrium with strictly
positive voluntary contributions, a generation optimally chooses its contribution, correctly
anticipating its return. The first order condition that determines the subscription T (x) in
state x is:

Eu a [x, x+ ] | x = Eu b [x, x+ ]r (x, x+ ),

14 GABRIELLE DEMANGE AND GUY LAROQUE

which, using the definition of mrs, can be rewritten as:


T (x) = γ (x+ ) mrs(x, x+ )T (x+ ).
x+ x

There is always an equilibrium without subscription: if I expect next generation not to con-
tribute, I shall not contribute anything myself. When there exists an equilibrium associated
with positive subscriptions T (x), the conditions of Theorem 1 are met, and the equilibrium
is interim optimal.12 Note that the pension fund return is equal to the rate of growth of the
population multiplied by the ratio of the next to current generations contributions.
In this section we investigate under which conditions this result generalizes to a hetero-
geneous population, when agents may only contribute to a collective fund. The functioning
of the economy is as follows. Let T i (x) be the subscriptions to the fund of a young agent
of type i born in state x, and r (x+ | x) the (rationally expected) return on subscriptions to
the social security fund when the next period state is x+ .

Definition 2: An equilibrium with collective social security is composed of an investment

and subscription decision (s i (x), T i (x)) for each agent i = 1, . . . , I, and each state x ∈ X,
all non negative real numbers, and a return r (x+ | x), for all (x, x+ ), such that:

1. for each agent i, and in each state x, (s i (x), T i (x)) maximize i’s expected utility in
state x, Eu i (a i , bi ) | x under the constraints
 i
a + s + T = w (x)
 i i i

b = ρ+ s + r (x+ | x)T i
i i

 i
s ≥ 0, T i ≥ 0.

2. The social security system has a balanced budget:

I
γ+i µi T i (x+ )
r (x+ | x) = i=1
I .
i=1 µi T i (x)

When there is no uncertainty at all nor production, we fall back on an economy à la

Samuelson with heterogeneous agents, in which, by stationarity, all strata grow at the same
rate. The contributions are constant for each strata and the return to the fund is equal to
the rate of growth of the population. One easily checks that an equilibrium with positive
contributions always exists and leads to a golden rule equilibrium. With uncertainty, a new
role is performed by a collective fund: it implies some (random) transfers between the
population strata.

Theorem 2: An equilibrium with strictly positive subscriptions to the collective pay as

you go system is interim Pareto optimal in either of the following cases:
(1) all populations have identical preferences and endowments, and only differ by their
growth rates,
(2) all populations have identical Cobb Douglas preferences.
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 15

This result shows that the collective system works optimally for sharing the demographic
risks when the agents have identical preferences and resources, as in (1), or in the case (2)
of Cobb Douglas utilities. This implies that the collective system yields interim optimality
in the example of Section 3, where preferences are Cobb Douglas, and this would be the
case even if aggregate uncertainty were allowed. Actually the collective system, under the
assumptions of Theorem 2, both equalizes the marginal rates of substitution of the living
young agents (intragenerational risk sharing) and satisfies the intergenerational optimality
condition. Full interim optimality is a strong requirement, and cannot be provided by a
collective fund in general: the assumptions are quite restrictive.
There is a natural way to disentangle intragenerational and intergenerational risk sharing
by having two kinds of institutions: financial markets to allow young agents to insure
themselves, pension funds to perform transfers between generations and to allocate the
demographic risks. More precisely Definition 2 may be adapted to the situation where each
agent in state x may buy (or sell) claims contingent on the next states x+ : the markets then
are sequentially complete. An equilibrium in this more general framework involves, on top
of the contributions and the return of the collective system, a collection of state prices and
individuals’ state claims for each future state conditional on the current state. In an economy
with a finite set of agents, the equilibrium would be optimal in such a circumstance. When
there is an infinity of agents as here, this is not the case any more, and the social security
fund (or equivalently a fixed quantity of an exogenous asset) is the appropriate instrument
to reestablish optimality.

Theorem 3: Assume sequentially complete markets: in each state x there is a complete set
of contingent markets on the immediately following states x+ . Then an equilibrium with
strictly positive subscriptions to the collective pay as you go system is interim Pareto optimal.

The proof is simple. Capital accumulation is clearly optimal and the equilibrium in the
state contingent markets equalize the marginal rates of substitution so that conditions (6)
and (7) are satisfied. As for intergenerational consumption sharing, remark that, using the
equality γ+i µi = γ+ µi+ , the return may be expressed as

γ+ T (x+ )
r (x+ | x) = ,
T (x)
I
where T (x) is the average contribution in state x: T (x) = i=1 µi T i (x). At an interior
point, the subscription
 of the typical agent i in state x is determined through the first order
condition:13 1 = x+ mrsi (x, x+ )r (x+ | x) for all x. Substituting for the return and using
the equality of the marginal rates of substitution yields

T (x) = γ (x+ ) mrs(x, x+ )T (x+ ),
x+

which can be rewritten as:

T = AT.
It follows that 1 is the largest eigenvalue of A and that (8) is satisfied. ✷
16 GABRIELLE DEMANGE AND GUY LAROQUE

The assumptions of Theorem 2 or 3 are quite restrictive. The crucial condition bears on
the optimality of intragenerational risk sharing. We now examine more general situations,
where agents have access to multiple funds and markets are incomplete: can the multiplicity
of funds be a substitute to missing markets?

5. Multiple funds

From now on, there may exist several funds, indexed by k = 1, . . . , K , which all offer a
pay-as-you-go retirement scheme. T i (x, k) is the subscription of a young agent i born in
state x to the kth social security fund (we allow an agent to subscribe to several funds). The
return of the kth fund is equal to14 :

i=1,...,I γ+ µ T (x + , k) T (x+ , k)
i i i
r k (x+ | x) =  => γ+ .
i=1,...,I µi T i (x, k) T (x, k)
Some funds may be reserved for some categories of populations, with constraints of the
form T i (., k) = 0, for some couples (i, k). This setup covers the situation where there
are specific funds associated with each population strata, as well as our previous collective
arrangement. The definition of an equilibrium in this more general framework follows from
a simple adaptation of Definition 2. Note that the number of funds in operation, as well as
their characteristics, are determined endogenously at an equilibrium. We typically expect
to see a multiplicity of equilibria, each one corresponding to a different structure of funds.

5.1. Sequentially complete markets

When there are complete markets, single vs. multiple funds perform equally well.

Theorem 4: Let markets be sequentially complete, and consider an equilibrium with

possibly multiple funds. If every strata subscribes positively to at least one fund in all
states, all the funds have the same rates of return: the same equilibrium could be obtained
via a unique collective fund.

Theorem 4 says that, if financial markets allow for optimal intragenerational risk sharing,
then a unique collective pay as you go system is enough to provide adequate intergenerational
transfers.
We can illustrate Theorem 4 with the previous example of Cobb Douglas preferences
a 1−β bβ , allowing for uncertainty in the aggregate wage. From Theorem 2, the equilibrium
of such an economy, with collective social security and absent financial markets, yields an
interim optimal allocation. Consider the system of multiple funds, one by category. Because
of Cobb Douglas preferences, in this particular case, the consumptions of the young agents
are equal to what was observed with a single fund, still given by15 a i (x) = (1 − β)wi (x).
With complete contingent markets all marginal utilities are proportional to the implicit
contingent prices, which gives here:

bi (x, x+ ) = α(x+ )(1 − β)wi (x)

SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 17

for some α(x+ ). Adding up and using the feasibility constraints give the value of α(x+ ):

βw(x+ ) = α(x+ )(1 − β)w(x).

Using this expression yields

w(x+ ) i
bi (x, x+ ) = β w (x)
w(x)

and we fall back on the allocation obtained through the collective system, without financial
markets. Note that complete markets play an important rôle for the determination of the
equilibrium with multiple funds. They allow the agents to modify their contributions to the
funds, as in the Modigliani Miller argument, so that the returns to the individual funds are
all equal in equilibrium, and equal to the return of the unique fund in the collective system.
We now turn to the more interesting case of incomplete markets.

5.2. Sequentially incomplete markets

When markets are incomplete, one could think that, in our setup of voluntary subscriptions,
the variety of funds might play the role of a variety of assets, and replace some of the
missing markets. Theorem 5 shows that this is not the case:

Theorem 5: If, at an equilibrium with multiple funds, a strata subscribes positively to

two different funds in all states, the two funds yield the same return and could be aggregated
without affecting the equilibrium.

Subscribing to several funds does not improve risk sharing at least if one sticks to the
natural assumption of positive subscriptions. It also implies that the coexistence of a col-
lective fund and a complementary fund is not justified if the contributions are based on a
voluntary basis. Note however that we do not investigate the situation where the positivity
constraint binds in some states for the subscriptions to some funds. It is possible that in
such a case the existence of several funds be a partial substitute for the absence of complete
markets for the exchange of risks.
Following Theorem 5, without loss of generality, we restrict ourselves to social security
systems in which each strata subscribes to a unique fund. Therefore the society is divided
into a partition of types, all the members of an element of the partition subscribing to the
same fund. We are now in a position to consider our initial question: are there cases where
a collective fund is superior to funds per category, and provides better insurance against
idiosyncratic shocks in the society? The following theorem gives a negative answer.

Theorem 6: Consider two social security systems with the same set of financial assets and
two corresponding equilibria. If, in each case, each population strata subscribes positively
to a fund, then the two equilibria are not comparable through the interim Pareto optimality
criterion.
18 GABRIELLE DEMANGE AND GUY LAROQUE

Theorem 6 means that a collective pay as you go social security scheme and special-
ized schemes by categories of agents are not comparable from the point of view of in-
terim Pareto optimality. Once each agent voluntarily subscribes to some pay-as-you-go
social security scheme, he is part to a set of intergenerational transfers, which seems to
be good enough as far as interim Pareto optimality is concerned. If there are individ-
ual social security funds, it is not possible to improve the allocation, i.e. to increase the
welfare of everybody in every state of nature when young, by introducing a collective
system.
It is useful to put in perspective these results with the example considered in Section 2.
The example has Cobb Douglas utility functions, so that collective social security brings
Pareto optimality, even in the absence of financial markets. From Theorem 3, we know that
with general utility functions sequentially complete markets and a collective social security
scheme still yield Pareto optimality. Specialized funds are equivalent to a collective fund.
Now in the second best situation where financial markets are incomplete, interim Pareto
optimality cannot be hoped for, but it might be the case that some organization dominates
the others. This is not the case: the collective and specialized social security funds are
always non comparable. If agents are allowed to vote with their feet and to build their
own pension fund for themselves, they would probably destroy a collective system, which,
contrary to our initial intuition, is unable to provide adequate insurance against idiosyncratic
shocks.

Appendix: Proofs

Proof of the necessary and sufficient condition for boundedness: Let s = (s(x))x∈X be
an accumulation strategy, and let:
γ (x)
F(s; x− , x) ≡ s(x− ) − s(x).
ρ(x)
Also, let C be the set of pairs (x− , x) in X × X , such that the probability of reaching state
x ¿from state x− is strictly positive. By (3), an accumulation strategy is feasible if and only
if:

s(x) ≥ 0 for all x ∈ X
γ (x)
 F(s; x− , x) ≥ − w(x)¯ for all (x− , x) ∈ C,
ρ(x)
and the consumption allocations are bounded if and only if F is bounded on the set of
feasible accumulation strategies. Now, since F is linear in s, F unbounded on the set of
feasible strategies is equivalent to saying that there exists s, positive different from 0, such
that:

F(s; x− , x) ≥ 0 for all (x− , x) ∈ C,

with at least one strict inequality for a couple (x− , x) in C. Take a cycle (x1 , ., xi , .., x1 ).
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 19

Then:

 γ (x2 )

 s1 ≥ s2 ,

 ρ(x2 )


 γ (xi )
si−1 ≥ si for i = 3, . . . , n (P0)

 ρ(xi )



 γ (x1 )

sn ≥ s1 .
ρ(x1 )
Assume that one of the stocks si is strictly positive in (P0). Then surely si−1 > 0, and
by induction along the cycle, all stocks are strictly positive. Multiplying term by term the
inequalities in (P0), and simplifying by the strictly positive product of the stocks yields

n
γ (xi )
P0 holds and si > 0 for at least one i ⇒ ≤1 and sj > 0. (P00)
i=1
ρ(xi )

If there is at least one strict inequality in (P0) then surely one stock is strictly positive.
Using a similar argument as above one gets


n
γ (xi )
P0 holds with a strict inequality ⇒ < 1. (P01)
i=1
ρ(xi )

Now under the assumption

F(s; x− , x) ≥ 0 for all (x− , x) ∈ C,

with at least one strict inequality for a couple (x− , x) in C, (P0) is met for all cycles with at
least one cycle with a strict inequality. For this cycle (P01) is met and stocks are strictly pos-
itive. Now consider all the cycles to which a state in the first cycle belongs. Surely all stocks
on these cycles are strictly positive. Continuing this way and using the recurrence property
shows that all the stocks are strictly positive and therefore (P00) is met for all the cycles.
We have shown that the set of feasible allocations is unbounded if and only if:


n
γ (xi )
≤1
i=1
ρ(xi )

for all cycles, with one strict inequality for at least one of them. Taking the contraposate
concludes the proof. ✷

To simplify notation we write


 ∂u i (a i , bi (x, x+ )) 
u ia [x, x+ ] = i i
∂a i a =a (x)


and similarly for u ib [x, x+ ].
20 GABRIELLE DEMANGE AND GUY LAROQUE


Proof of Theorem 1: An optimal allocation maximizes i,x λi (x)E{u i [a i (x), bi (x, x+ )]
| x} under the feasibility constraints


I
µi (x)γ (x){[a i + s − wi ](x) + bi (x− , x)} − ρs(x− ) = 0, for all (x− , x).
i=1

Note that, we have rewritten (3) by using (1), i.e. µi (x)γ (x) = µi (x− )γ i (x). The program
is concave so that the first order conditions are necessary and sufficient for optimality.
(6) comes directly from the optimization with respect to s(x). Let φ(x− , x)/γ (x) be the
multipliers associated with the feasibility constraints. The derivation of the Lagrangian with
respect to a i (x) and bi (x, x+ ) yields:
 i   
λi (x)E u  a [x, x+ ]  x = µi (x)γ (x) φ(x− , x), (P1)
x−

and λi (x)u  ib [x, x+ ] Pr(x+ | x) = µi (x)φ(x, x+ ), which may also be written as:
 i  
λi (x) mrsi [x, x+ ]E u  a [x, x+ ]  x = µi (x)φ(x, x+ ). (P2)

P1 and P2 immediately yield that the marginal rates of substitution are independent of i
so that (7) is met. Moreover the vector V defined by

λi (x)Eu  ia [x, x+ ] | x
V (x) = ,
µi (x)

is equal to γ (x) x− φ(x− , x), and is independent of i as well. Plugging in the expressions
of φ(x− , x) as given by P2 yield

 λi (x− )Eu i [x− , x] | x−

V (x) = a
γ (x) mrsi [x− , x] for all x
x− µi (x− )

or finally since A(x− , x) = γ (x) mrsi [x− , x]


V (x) = V (x− )A(x− , x) for all x.
x−

Therefore V is a positive left eigenvector associated with the eigenvalue 1 of A, which

implies (8).
Conversely assume that (7) and (8) are met. We may find weights and multipliers so that
the first order conditions P1 and P2 are met as follows. If (8) holds, choose V a positive
left eigenvector associated with the eigenvalue 1 of A and define λi (x) such that

λi (x)Eu ia [x, x+ ] | x
= V (x).
µi (x)
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 21

Defining

φ(x− , x) = V (x− )A(x− , x)

and using that by (7) A(x− , x) = γ (x− ) mrs

 [x− , x] immediately gives P2. Now the equality
i

characterizing the eigenvector: V (x) = x− V (x− )A(x− , x) for all x yields P1. ✷

Proof of Theorem 2: We first show that under the assumptions the marginal rates of
substitution mrsi (x, x+ ) of different agents i are equal so that condition (7) is met. In case (1)
this is trivial since the agents have the same preferences and face the same budget constraints
at both periods so that their consumptions are identical. In case (2) they face the same
investment opportunity set but differ in their first period income. However with identical
Cobb Douglas preferences, it is well known that the first period decisions, consumption and
investment, are proportional to income. Therefore all young agents consumptions at both
periods are proportional and the marginal rates of substitution are equal.
We may now drop the index i for the marginal rates of substitution and show that (8) is
met. The first order conditions that characterize the solution of the consumer program are:

mrs(x, x+ )ρ(x+ ) ≤ 1, with equality if s i > 0,
x+

so that condition (6) is satisfied at an equilibrium and


mrs(x, x+ )r (x+ | x) ≤ 1, with equality if T i > 0.
x+

The last condition can be expressed in terms of the matrix A introduced in (8), once one
substitutes r (x+ | x) with its expression in terms of the transfers. It follows that the first order
condition associated with the contribution to the social security system of the population
strata i can be written:

A(x, x+ )T (x+ ) ≥ T (x), with equality if T i (x) > 0. ✷
x+

Proof of Theorem 3: Since contingent markets are complete the marginal rates of sub-
stitution mrsi (x, x+ ) of different agents i are equal and then the proof is the same as that of
Theorem 2. ✷

Proof of Theorem 4: We first show that at an equilibrium of a social security system in

which a strata i subscribes positively to a given fund k in all states, then Ai T k = T k where
the matrix Ai is defined by Ai (x, x+ ) = γ (x+ ) mrsi (x, x+ ). It says that the positive vector
of the average contributions to the fund T k is a positive eigenvector of the positive matrix
Ai , implying that Ai has its largest eigenvalue equal to 1.
The first order conditions that characterize the solution of the consumer program are:

mrsi (x, x+ )r k (x+ | x) | x = 1 for all x.
x+
22 GABRIELLE DEMANGE AND GUY LAROQUE

Substituting with the expression of the return of the fund yields


γ (x+ ) mrsi (x, x+ )T (x+ , k) | x = T (x, k) for all x.
x+

which can be expressed as

Ai T k = T k (P3)
We now prove Theorem 4. Since markets are complete the marginal rates of substitution
are equalized through the agents and all the matrices Ai are equal to some A. The positive
vector of the average contributions to any fund k is an eigenvector of A associated to its
largest eigenvalue. By a result on positive matrices the associated eigenspace is of dimension
1 so that the vectors of the average contributions to the different funds are proportional. It
follows that the returns of all funds are identical: a unique collective fund with the same
individual contributions would yield the same equilibrium allocation. ✷

Proof of Theorem 5: The proof follows from (P3). The positive matrix Ai has a unique
positive eigenvector vector, up to a scalar. Therefore if i contributes to two funds their
average contributions are proportional and the returns are equal: they can be aggregated.
✷

Proof of Theorem 6: Consider two equilibria in which i contributes to fund k or k ∗

respectively. Let [s i (.), T i (., k)] and [s ∗i (.), T i (., k ∗ )] be the associated decisions. Note
that for any λ in [0, 1], the linear combination of the two equilibria allocations with weights
respectively equal to λ and (1 − λ) is feasible, and is obtained through a consumer investment
policy equal to [λs i (.) + (1 − λ)s ∗i (.), λT i (., k), (1 − λ)T i (., k ∗ )], where the funds k and
k ∗ have the same return as in the two equilibria. Let V i (λ, x) be the utility of the consumer
born in state x at the λ allocation. The V i ’s are concave in λ. The proof of Theorem 5 consists
in showing that the derivative of V i (λ, x) with respect to λ at the point λ = 0 cannot be
positive for all x, with a component strictly positive for some x.
We have:

dV i (λ, x)   i 
 = E u  a [x, x+ ] | x {s i (x) + T i (x, k) − s ∗i (x) − T i (x, k ∗ )}
dλ λ=0
 i
− u  b [x, x+ ] Pr(x+ | x) {ρ(x+ )[s i (x) − s i∗ (x)]
x+

+ r k (x+ | x)T i (x, k) − r k∗ (x+ | x)T i (x, k ∗ )},

where the marginal utilities are computed at the original equilibrium allocation. By the first
order conditions at equilibrium, the above equality reduces to:

d V i (λ, x)   i 
 = −E u  a [x, x+ ] | x {s ∗i (x) + T i (x, k ∗ )}
dλ λ=0
 i
+ u  b [x, x+ ] Pr(x+ | x) {ρ(x+ )s ∗i (x) + r k∗ (x+ | x)T i (x, k ∗ )}.
x+
SOCIAL SECURITY WITH HETEROGENEOUS POPULATIONS 23

Now, by the first order condition associated to the optimal investment (i.e. (6)), the contri-
bution of s ∗i (x) to the derivative is non positive. The factor which multiplies T i (x, k ∗ ) is
equal to, up to a positive proportionality factor:

 T (x+ , k ∗ )
Ai (x, x+ ) − 1.
x+ T (x, k ∗ )

If the star equilibrium is preferred for all x, then:

Ai T ∗ ≥ T ∗ ,

which is only possible if T ∗ is proportional to T . ✷

Notes

1. Under the name of conditional optimality.

2. One may prefer to derive these formula from the population levels. With obvious notation, one has
I
i=1 N− γ
i i
Ni Ni
γi = , γ =  and µi =  I .
N−i I i
i=1 N− i=1 Ni

3. Recall that, by Note (1), the knowledge of x yields that of the proportions of old agents in the economy.
4. For generic values of the growth rates, the composition of the population by type takes an infinite number of
values, and our analysis would need to be generalized to an infinite X . We believe that the main results carry
over to this case, as long as X is compact.
5. Note that we have assumed that the rate of return is a macroeconomic variable, identical for all the participants
in the economy: for a given generation, there is no scope for insurance against this macroeconomic shock.
6. With simplified notation, the first order condition is u a = E(u b r̃ ) where r̃ is the expected return of the fund
at the time of the decision. Using the fact that

u b βa
= , b = (w − a)r̃ ,
u a (1 − β)b

one gets (1 − β)(w − a) = βa.

7. The same allocations would result independently of the utility functions if the contributions were mandatory,
proportional to the wages as is often the case in practice.
8. Utility levels have been divided by K = (1 − β)(1−β) β β .
9. We avoid technical complications by only considering strictly positive weights.
10. See the Appendix for a proof.
11. A characterization of optimality for economies with multiple goods and non stationary allocations is available
in Okuno-Zilcha [1980], using spot prices.
12. This result is proved in Demange-Laroque [2000] in a more general set up with endogenous (possibly in
infinite number) states. However such an equilibrium may not exist: existence depends on the value of an
eigenvalue of the matrix A computed at the ‘autarkic’ equilibrium, which is obtained when there are no
intergenerational transfers, T = 0. When the above equilibrium does not exist, there is an equilibrium with
negative contributions which is interim optimal. In such a case, all the young agents receive transfers from
the old ones in all the states.
13. Eu ia [x, x+ ] | x = Eu ib [x, x+ ]r (x+ | x) | x.
 
14. Again, the equality γ+i µi = γ+ µi+ gives i γ+i µi T i (x+ , k) = i γ+ µi+ T i (x+ , k) = γ+ T (x+ , k).
24 GABRIELLE DEMANGE AND GUY LAROQUE

15. The first order conditions are u a = E(u b r̃k ) where r̃k is the expected return of asset k. Therefore the aggregate
pensions to be distributed are still equal to βw(x+ ). Denoting by sk the amount invested in asset k and using
u b /u a = (βa)/((1 − β)b) this gives

βa 
E r̃k = 1 with b = sk r̃k .
(1 − β)b k

Note that multiplying the above equations
 by sk and summing up over all k yields (1 − β)( k sk ) = βa,
which gives a = (1 − β)w since k sk + a = w.

References

AIYAGARI, S.R. and PELED, D. [1991]: “Dominant Root Characterization of Pareto Optimality and the Existence
of Monetary Equilibria in Stochastic Overlapping Generations Models,” Journal of Economic Theory, 54, 69–83.
DEMANGE, G. and LAROQUE, G. [1999]: “Social Security and Demographic Shocks,” Econometrica, 67(3),
527–542.
DEMANGE, G. and LAROQUE, G. [2000]: “Social Security, Optimality and Equilibria in a Stochastic Overlap-
ping Generations Economy,” Journal of Public Economic Theory, 2(1), 1–23.
EDWARDS, S. [1996]: “The Chilean Pension Reform: A Pioneering Program,” NBER Working Paper # 5811.
FELDSTEIN, M. [1995]: “Fiscal Policies, Capital Formation, and Capitalism,” The Joseph Schumpeter Lecture
of the European Economic Association, European Economic Review, 39, 399–420.
GALE, D. [1973]: “Pure Exchange Equilibrium of Dynamic Economic Models,” Journal of Economic Theory,
6(1), 12–36.
MANUELLI, R. [1990]: “Existence and Optimality of Currency Equilibrium in Stochastic Overlapping Genera-
tions Models: The Pure Endowment Case,” Journal of Economic Theory, 51, 268–294.
OKUNO, M. and ZILCHA, I. [1980]: “On the Efficiency of a Competitive Equilibrium in Infinite Horizon Monetary
Economies,” Review of Economic Studies, 47(4), 797–807.
PELED, D. [1984]: “Stationary Pareto Optimality of Stochastic Asset Equilibria with Overlapping Generations,”
Journal of Economic Theory, 34, 396–403.