Journal of Financial Economics I (1971) 225-244.

Q North-Holland Publishing Company

AN AGGREGATION THEOREM FOR SECURJTIES MARKETS*

Mark RUBINSTEIN
Graduate School of Business Adminisrrarion, Unirersity of Californiu,
Berkeley. Cali/. 94720. U.S.A.

Received December 1973, revised version received February 1974

Alternative sets of sufficient conditions are developed under which equilibrium security rates
of return are determined as if there exist only identical individuals whose resources, beliefs,
and tastes are a composite of the actual individuals in the economy. These conditions include
as special cases all those previously examined in the literature (including conditions sufficient
to produce the two-parameter mean-variance model), as well as others. Whenever such a
composite individual exists it is shown that (I) valuation equations take a specific form and
contain only exogenous parameters of the economy; (2) market exchange arrangements are
Pareto-optimal; and (3) competitive value-maximizing firms make completely specified
Pareto-optimal production decisions both over dates and states. These results rely on the
observation that under popular homogcncity assumptions regarding beliefs and tastes, even
though the securities market may be incomplctc. equilibrium rates of return are determined
as if there were an otherwise similar Arrow-Dcbreu economy.

1. Introduction

The chief difftculty befouling the analysis of sccuritics market equilibrium

is the problem of aggregation. When individual decisions arc aggregated to
form equilibrium relationships, individual-specific choice information must bc
eliminated and replaced by cxogcnous parameters of the economy. Solution
of the aggregation problem then leads to closed-form valuation equations and
thcorcms on the optimality of market exchange arrangements and production
decisions.
This paper demonstrates that valuation equations and optimality theorems
are quite easy to derive if the aggregation problem can bc solved first. In parti-
cular, it is shown that whenever an individual can be constructed whose rc-
sources, beliefs, and tastes are a composite of the actual individuals in the
economy, then (I) valuation equations take a specific form and contain only
exogenous parameters of the economy (section 4); (2) market exchange arrange-
ments are Pareto-optimal (section 5); and (3) competitive value-maximizing
firms make completely specified Pareto-optimal production decisions both
over dates and states (section 6). Section 2 describes the economy, and section
3 develops alternative sets of sufficient conditions under which a composite
individual may be constructed. These include as special cases almost all published

*Research for this paper was supportedin part by a grant from the Dean Witter Foundation.
226 M. Rubinstein, Aggregation theorem for securities markets

conditions for which the aggregation problem has been solved, as well as others.’
The aggregation theorem relies on the observation that under certain con-
ditions, even though the securities market may be incomplete, the equilibrium
rates of return will be determined as if there were an otherwise similar complete
(Arrow-Debreu) market. For example, these conditions include those sufficient
to produce the two-parameter mean-variance equilibrium model. Therefore, in
a fundamental sense, these incomplete market models are a special case of an
Arrow-Debreu economy. When viewed in this light, it is not surprising that the
optimality features of the Arrow-Debreu economy carry over to an appro-
priately specified incomplete market.

2. The economy

2. I. General markers

Each individual in a perfect and competitive economy allocates his present

wealth bk’, among present consumption C, and the remainder W,,-C,, to
risk-free and many risky securities. a denotes the proportion of W,-Co he
allocates to risk-free securities, which earn rate of return rF. J?, denotes the
proportions of the remainder (W, - C,,) (I -a) he allocates to each of J (j = I,
2 . .,J) risky securities so that &pi = I. Each security j earns rate of return
r;c’ if each of E (E = I, 2, . . ., E) states occurs. Consequently, if state e occurs
his future wcnlth

wlc = (wO-CO)(I +ar,+(l -a) C bjf],).

Each individual is assumed to obey the Savage axioms of rational choice:

Hc has beliefs and tastes reprcscntable by probabilities and a utility function
so that he maximizes his expected utility. Let n, dcnotc the subjective probability
hc attaches to state e and let U(C,)+/,V( W,,) denote his additive utility function
over present consumption and future wealth, where p is a constant denoting
his rate of patience and U and V are functions. Assume U’ > 0, U” < 0,
p > 0, V’ > 0, and V” < 0.
In brief, the individual solves the following programming problem:

max UCo)+P 1c n,V[(kf’o-Co)(l +ar,+(l -a) 1 /Ifi,)]-A[ Cpj- I],

co. a. IPI1 I I

‘For example. Mossin (1973, pp. 68-73) has solved the aggregation problem under con-
ditions of homogeneous beliefs and quadratic utility and Linfner (1969) under conditions of
exponential utility and normal probability asscssmcnfs. Wilson’s (1968) classic paper ‘The
theory of syndicates’ derives these as well as other solutions causing section 3 of this paper
to be somewhat redundant. However, by way of contrast. the aggregation theorem of this
paper is cast within an explicit securities market equilibrium context, considers choice of both
present consumption and fulurc wealth, and is dcmonstratcd by more transparent proofs.
See also Wihon (1967) for a muhiperiod analysis of the aggregation problem under conditions
of limeadditive exponential utility. Hakansson (1970) also has developed results that overlap
IO some extent those of sections 3 and 5 of this paper.
 M. Rubinstcin. Aggregation theorem for secwities markets 227

where 1 is a Lagrangian multiplier. Differentiating partially with respect to each

choice variable, the necessary and sufficient conditions for an optimum are

U’(G) = (1 +rM X,YWk)l.1 (1)

I?
C n.V’(bV,e)(rj,-r,) = 09 (alli), (2)
;I. = (wO-C,)(l +ar,+(l-a) C fijrjc)v where c/I, = 1,
i i
(all e). (3)

To provide for the endogenous determination of security rates of return,

we append closure conditions that effectively take production decisions across
dates and states as given. Let Ci’and {Sjrc) be respectively the aggregate present
consumption and the aggregate future dollar return for each of the J risky
securities if state e occurs. Consequently, if the population of the economy is
f(i= I.2 ,..., f).then
7 cb = c,“, (4)

T (Wg-c;)(l +r,)a’ = 0, (5)

~(~~-C~)(l+f,=)(l-~l~)B; = Sjl~. (alljande). (6)

In equilibrium, these three closure conditions must hold for the optimum
choices of (CA. a’, {/l:]) for all i, along with conditions (l)-(3) which must hold for
each individual i.

2.2. ScporaIion

In general, each decision variable at the individual lcvcl will depend on all
paramctcrs of the model (We, rr, (r,,.}, {n,}, U, p, V). Conscqucntly, the con-
sumption C,, lcvcragc a and diversilication (/Ij} decisions must be made
simultaneously. 1lowcver, thcrc is a class of utility functions V( IV,,) for which
thcsc decisions arc to some extent scparablc. In particular, consider the class
of utility functions satisfying the differcntinl equation
-V’(W,,)/V”(CV,J = A+BW,,, (7)
H here A and B arc fixed parameters. This equation has three solutions’ depcnd-
ing on the value of B.

(3) UW,,)- -A cxp (- W,JA), (B = 01,

(b) VW,.bWf+ WIJr (B = 11,

(cl V~V,.)~ &A+ BWre)l-b, (B # 0, I),

*The solutions lis~cd arc those possibly consistent with Y’ > 0 and Y’ c 0. Required for
this consistency in all cases A 2 0. Morcovcr. for lunctions (a). A > 0. and for functions
(c). if A = 0. then B > 0 and if B = - I, - 1. -1, . . ., then -A/R > WI..
‘28 M. Rubinstein. Aggregation theorem for securities markets

whereb rB-’ and - means ‘is equivalent up to an increasing linear trans-

formation to’. This class of utility functions is quite rich, containing as special
cases: the constant absolute risk aversion function (B = 0), the constant
proportional risk aversion function (A = 0), quadratic utility (B = -I), and
a cubic utility function (B = -f).
Although other separation results for this class of utility functions can be
derived,’ we shall only need the following theorem:

Theorem (Universal portfolio separation). If tize economy is composed of

individuals with the same beliefs and taste parameters 8. then the optimal portfolio
of risky securities {/Ii} will be the same for all individuals.

ProoJ Since elements of this proof have appeared in Pye (1967). only an
outline will be given. When B # 0, V’(W,J = (A+BW,J-* so that from
conditions (2) and (3)

~n,(A+R(W,-C,)(l+arF+(I-~)~~jr,,)]-b(rj,-rf) = 0, (all j).

. i

For a second individual (reprcsentcd by primed variables) faced with the same
opportunitics(r,, {rip}), suppose that {II; = n,} and B’ = E. Then similarly.

c x,[A’+B(E’A-Cb)(I +a’r,+(l -a’) 1 &r,,)]-*(rjr-rF) = 0,

c i
(all j).

The portfolio decisions of thcsc two individuals arc related by

1
A’+ B(Wh-C&l +r,)
(U$- CJC I - a’)p; = ---. .- _
A+B(IV,-C,)(l+rF)
(alI j).
x ( I+‘, - CON1-a)/],,
:ince this equation satisfies the above first-order conditions. Summing this
equation over all j and since Cl/?; = x,/I, = I,

Dividing this into the prior equation, S; = /I, for all j, When B = 0, then
V’( iv,,) = exp (- lV,JA). A similar analysis but using

‘For example, it can be shown that: (1) If the economy is composed of individuals with the
same beliefs and haste pnramcters A = 0 and Il. then the optimal leverage c~and diversification
(J?;) decisions will be the same for all indivitlualc (even though they have difierent resources.
rates of patience, and utility functions U). (2) If, for any individual, A = 0 and B = 1 (loga-
rithmic utility). then his optimal consumption decision CO is independent of his opportunities
(r, , {I,.)) and beliefs.
 .%I. Rubinsrein, Aggregation theorem for securities markers 2’ 9

( Wh - Ch)( I - Cf’)/l; = [A’/A]( IV, - C~J)(

1- Cf)/3j

will also yield the result /I> = /Ii for allj. Q.E.D.

Let r MP G xj/?jrjt denote the rate of return on this common portfolio of

risky securities. Since all risky securities must be held, III can be interpreted
as the market portfolio of all securities. All individuals will hold this same
portfolio of risky securities even though they have different parameters
(W,,, U, p and A). Observe that this conclusion is independent of the number
J of risky securities available in the market. In particular, whether the number
of securities is less than the number of states (J+ 1 < E) or equal to the number
of states (J+ I = E), the theorem holds.

2.3. Conrplete markets

In general. J+ I < E. However, it will be useful to consider the simplification
in conditions (l)-(6) that results under a complete securities market (J+ 1 = E).
In this case, it is well known that choices will be made as if there exist a full
set of ‘state contingent claims’ in place of the actual securities. That is, indivi-
duals behave as if for each state e there exists a security for which

rr > -1, if e occurs,

r. = -I, otherwise.

With a one-to-one corrcspondcncc bctwocn sccuritics and states. securities

may be indcxcd simply by subscript U.
In this cast the wealth constraint rcduccs to W = ( lVO-CO)(I +r,)/?,
(all e), so that the programming problem bccomcs

max ~(C,)+P C n,k%W~-C,)(l +r,V41-4 2 A- Il.

co.16.1 e l

with necessary and sufftcicnt conditions for an optimum

(I’(C,) = (I+ rr)PXr V’( W,e)r (all e),

w,, = ( IV, - C,)( I+ r,Mi 9 where ; p, = I, (all e).

Ilowevcr, since given choice of C, there is a one-to-one correspondence

bctwcen choice of /I, and IV,=, the problem can be rcformulatcd as one of
choosing (IV,,} directly. Summing the wealth constraint over states c and since
J5P. = 1,
N’, = co+ 1 IV,,(l +rJ-‘.
c
The reformulated programming problem becomes

max I/(C,)+p~~=Y(W,,)-I[C,+ 2 W,,(I+r,)-‘-Wol,

co.IWI.1 c l
 230 M. Rubinstcin. Aggregufion rheorem for securities markets

with necessary and sufficient conditions for an optimum

~‘(CCB) = (1 +r,)PLUW,,), (all e) (8)

IV, = co+ c W,,(I+r,)-‘. (9)

c

Closure conditions replacing (4)-(6) are simply

c: = cc;, (10)
I

Since they permit incomplete markets, economies with universal portfolio

separation would appear to be a generalization of economies with complete
markets. However, this is deceiving in the following sense: in an econon~y with
universal porffolio separation in&duals reach the same optimal allocation in
terms of final outcomes (CA, { Wf,}) as in the corresponding complete markets
economy with the same homogeneity assumptions. For example, if all individuals
have homogeneous beliefs and quadratic utility (II = -I), then the same
optimal allocations (CA, { Wf,}) arc reached in an otherwise similar economy
but with complctc markets. An immediate conscqucncc of this is that th rutcs
of return on securities in an cconorrry with unircrsal port/;&o separation but with
incomplete markets arc cit~termincd as IY there were complete markets. (See
appendix 2.) This result will prove useful throughout this paper.

3. Homogeneity and aggregation

With this preparation, ~c arc ready to dcvclop the main thcorcm. I:or the
economy dcscribcd in section 2, the aggregation problem is the derivation of
rates of return (TV, {rj,.‘,) in terms of the exogenous spccifications4 of the cco-
nomy: beliefs (nt}, tastes {I),. Ui. Vi]. production (Ci’. {Sj,,}), and population I.
To solve the aggregation problem, we usually try to introduce production
variables into conditions (l)-(3) by summing them, rcspectivcly, over all
individuals:

1 C d v;( wfe)(rje- rF) = Ot (all j), (2’)

i c

wi”c = 1 sjlct (all f,). (3’)

I

*Generally, these exogenous specifications should also include thr initial endowment of
each individual’s claims to present consumption and future wealth. Houevcr, as a property
of all the solutions to the nggrcyation problem dcvrlopcd in this paper. equilibrium rates of
return arc inscnsitivc to rcdislribution of initial wealth. Conscqucntly. individual endowments
can be ignored.
 M. Rubinstein. Aggregarion theorem for securities markers ‘31

Eq. (3’) follows from (5) and (6). Unfortunately, (1’) and (2’) have remained
complex expressions, containing personal information that is the result of
individual choices (CA, { W’,,}).If there were some way to eliminate this indivi-
dual-specific choice information from conditions (1’) and (2’), the aggregation
problem could be solved.
We will solve the problem by assuming individuals have partially similar
economic characteristics. In the trivial case of identical individuals, since at
the optimum Ci = Cr/f = C,, and I+‘;, = WE/f E W,, for all i and e, (1’)
and (2’) reduce to

U’(G) = (I+ TF)P[ 1 % V’( W,,)l.

c

1 ncY’( by,.J(rj.-fF) = O, WA,

c

where the i differentiation can be omitted. Since together with (3’) these simpli-
fied conditions determine solutions to equilibrium rates of return (rF, {rj,.}),
and since the conditions arc specified in terms of exogenous parameters only,
the aggregation problem has been solved. Moreover, we can trivially construct
a composite individual with resources W, = W,“‘/f z xiWhjf, beliefs {n,}
and tasks p, U and V so that equilibrium rates of return are dctermincd as if
thcrc exist only these composite individuals.
Less trivial solutions to the aggregation problem can be obtained by rcstrict-
ing tastes to satisfy cq. (7). Moreover. since the equilibrium risk-free rate of
return must also bc determined, utility functions U and Y will bc ;ISSUIIRXI of
the same form so that

- U’(C”,/U”(C”) = A + BC,.

Even though individuals are allowed to differ in certain rcspccts, the aggrc-
gation problem can be solved. I-or example, as long as individuals have the
same beliefs, rates of patience, and taste paramctcrs U Z 0. they can have
dilferent resources and taste parameters Ai. Morcovcr, when B = 0 (exponential
utility), cvcn the rates of patience may direr. When ,4 = 0 and B = I (loga-
rithmic utility), the rates of patience may also differ provided individuals have
the same resources and bclicfs. The chief objection to these solutions is the
required homogeneity of beliefs. However. when B = 0 and a complete markeL
exists, then the aggregation problem may be solved even though individuals
have dilfercnt resources, beliefs. rates of patience and taste parameters Ai.
Also, when A = 0, B = 1 and a complete market exist>, then the :lggrcg:l-
tion problem may bc solved even though individuals have difkrent beliefs
provided they ha1.c the same resources and rate!, of patience.
The following theorem summarizes these remarks. A proof is provided in
appendix I.
 232 M. Rubinsrein, Aggregation theorem /or securities markets

Theorem (Aggregation). Consider the following sets of homogeneity con-

ditions: ’

(i) All individuals have the same resources W,, , beliefs {n,}, and tastes p, CTand V.
(ii) AII individuals hare the same beliefs {n,}. ra:es of patience p, and taste
parameters B # 0.
(iii) AN individuals have the same beliefs {n,} and taste parameters B = 0.
(iv) All individuals have the same resources W,, , beliefs {n,}, and taste parameters
A=OandB= I.
(v) A complete market exists and all individuals ha1.e the same taste parameter
B = 0.
(vi) A complete market exists and (111individuals hare the same resources W,
andtastesp, A = OandB = I.
Equilibrium rates of return are determined in case (i) as if there exist only com-
posite individuals each with resources W, . bclicfs (n,), and tastes p, U and V;
and equilibrium rates of return are dcternrincd in cases (ii)- as if there exist
only composite individuals each with the following economic characteristics:6

Rcsourccs w, = c Wi/I;
i

‘The suliicient conditions for the aggregation thcorcm permit some wcakcning. In cases
(ii)-( utility functions U and Y can dilfcr so that they satisfy

- U’(C,)/U”(C,) = A, + BC,, a11d

-V’(W,,)IV”(W,,) = A,+BW,,,
whcrc generally A, # A, . The composite parameters will then be A,, = E,AJ/, A, = Z, A,,//
and II. f Ioucvcr, when B = 0 thcrc must exist a constant k, the same for all individuals, such
that A,, = &A,, for all i. These generalizations permit a multipcriod cxtcnsion of the theorem
to the scenario in which individuals maximile

~~,$K)I~

whcrc E is an cxpcctation operator and T (I = 0. I. 2. . . ., T)is an individual’s lifetime. See

appendix 2 for wcakcning the complete market assumption of cases (v) and (vi).

When A #O and bclicfs arc hctcrogcneous. one cavca~ is in order. The composite beliefs
71, E nn;AI:IIAI,
W 4,
1
a geometric average of individual beliefs. does not fulfill all the properties of a probability
mcasurc. Although n. is non-negative for all c. Ln. # 1. cvcn though Ln.’ = I for all i.
 M, Rubinsrein,Aggregation theorem for securities markets 233

Tastes p E n piA-,, (A # 0).

L
Or

(1 +p)-’ = T(l +pJ-I/f, (A = O),

In brief, whenever the conditions of the aggregation theorem are met, equi-
librium rates of return are determined as if there exist only composite individuals,
each of whom solves the following programming problem :

max U(C,)
co. a, (#,I
+P 1 n,U(Wo-C,)(1+ ar,+(l -a) C Bjrj,)l-4 Bj- 11,
c j T
with necessary and sufficient conditions for an optimum

U’(G) = (1 +r&[ c n,U~Y,,)l, (12)

c
C ncY’( bV,c)(rjc-r,) = 09 (all j). (13)
c
Here C, E Ct/f, CYlc s K’:l/r = CjSjl,/I (all e), W, E W,“‘/l and {n,}, CJ,
V and p are appropriately defined composites.
Since equilibrium rates of return are uniquely determined by (12) and (13).
it follows trivially that:

Corollary (Rcsourcc distribution irrclcvancy). Whwcwr a composirc individual

can bc cons~ruc~cd. in equilibrium, rarer of rcfurn arc insetrsifivc to the clktribution
of rcsourccs among inriit~idual.s.
Conscqucntly. if the allocation of resources is Pareto-optimal (as will be shown
in sections 5 and 6) and if every Pareto-optimal allocation of resources can
bc achieved by an appropriate redistribution of resources. then any inter-
ferencc by a central planner that aitcrs rates of return must ipso facro lead to
a nonoptimal allocation of resources.
It also follows trivially from (12) and (13) that:
Corollwy (Population irrelevancy). Whcnccer a composire ittrlicirlual cat1 be
cottswrrctcd, in equilibrium, rates of return are ittsensi~ive to the population as
long us ~lrc economic churacteristics of the compasirc itrtlir~ichralremain uttchan~gctl.
For cxnmplc. in an economy of identical individuals, rates of return arc inscn-
sitivc to the population. Marc gcnerully, whcncver a composite individual
exists, rates of return arc insensitive to dccrctiscs or incrcabcb in the numhcr
of those same composite individuals.
The aggregation theorem provides only sets of sufficient conditions, not also
necessary conditions, for the construction of a composite individual. More
gcncrally, we will dcfinc a composite individual by the following traits:
234 M. Rubinstrin. Aggregation theorem for securities markets

(1) his initial wealth, optimal present consumption, and optimal future wealth
for every state are arithmetic averages of their corresponding aggregate
values;
(2) any homogeneous economic characteristic shared by all actual individuals
is also an economic characteristic of the composite individual;
(3) his beliefs for each state are a function of the beliefs of all actual individuals
for the corresponding state and at most also depend on the tastes of all
actual individuals;
(4) his rate of patience and taste parameters are, respectively, a function of the
rates of patience and taste parameters of all actual individuals and at most
also depend on the beliefs of all actual individuals;
(5) equilibrium rates of return are determined as if there exist only composite
individuals.

(I) imposes the requirement that IV,, E W,“/l, C,, s Cr/f and W,, E H’ri/f.
(2), for example, requires that if n: is the same for all i, then the composite
belief n, = nk. Observe that this is satisfied by

(3) and (4) exclude the construction of composite beliefs and tastes as functions
of initial wealth (IV;} and aggregate production variables CC,“, {S,,,}). When
production choice is introduced (section 6). it will bc shown that production
choices (Ci’, IS,,,}) by compctitivc value-maximizing firms, as well as equili-
brium rates of return (rF, {rj,}), arc dctcrmincd as if thcrc exist only composite
individuals. In brief, a composite (average) individual must bc arithmetically
average with rcspcct to cndogcnously detcrmincd variables, sham commonly
held traits, have traits corresponding to exogenous traits of actual individuals
that depend only on those cxogcnous traits and arc non-dictatorial, and make
the same decisions with regard to commonly shared endogenously dctcrmincd
variables as reached jointly by actual individuals.’

‘The composite beliefs of this paper should not be confused with the ‘consensus beliefs’
defined in Kubinstein (I973d). Consensus beliefs are those beliefs ~IIAI if held bv all individuals
in an otherwise similar economy would generate the same equilibrium rates of return. For
example, if a complete market exists and all individuals have the same tas~cs p, A = 0 and
B = 1. then consensus beliefs

n.” = z - IV,’ n.‘. (all e).

i lV”M
Since n:” is a function not only ol’ individual exogenous paranislsrv tn.‘} but also (IV,‘). which
are functions of the endogenously determined security rates of return. II,” is not a composite.
Therrfore, composite bchrfs are always consensus beliefs but not vice versa. This definition
also excludes other SCIS of homogeneity conditions. consistent uifh hetcrogencous normal
probability as>essnlcnts, from which Lintner (1969. 1972) has constructed complex ‘com-
posites’. These ‘composite’ parameters, to the extent they do not overlap cases (i)-(vi), are
all functions of aggregate production variables (S,,.}.
 M. Rubinsrrin. Aggregation theorem for securities markets 235

The remainder of this paper will work directly with (12) and (13) presuming
sufficient conditions for the construction of composite parameters are met.

4. Valuation

With the aggregation problem solveJ, it lernAn> !(I derive closed-form solu-
tions to the equilibrium rates of return.

Theoren~ (valuation’). W%cnc~er ;Y)W~OUI~ diciduaf can be constructed,

in equilibrium

(14)

lY(rj) = rF + Ah.(rj, - V’( W , ,) Std ri, (all j), (15)

where

A = Std [ P”( W’,)] E[ I-‘( Ii’,)] > 0.

Proof Eq. (14) folln~ immediately from (12). Solving (13) for theequilibrium
expcctcd rate of return on any risky sccurityj,

E(rj -rf)El(V(W,)]~t Cov [r, -rF, V’(W,)] = 0,

so that

E(rj) = r,+ LjV’( IV,)]- ’ Cov (r,,- C”(It’,)).

Eq. (!Z) follows immediately from the definition of tlic correlation coctficicnt.
Since V’(CV,) > 0, then I. z 0. Q.E.D.
Gcncr;llizing the discussion in Ruhinstcin (1972b, pp. i6S-17l), fc(rj, -. V’
( W,)) c;ln lx intcrprctcd as a mcasurc of the ,ron-~li~,crsi/rob~~,risk of security j,
It follows from (15) that the risk premium E(r,) -rF will bc positrvc if and
only if h.(r,, - Y’(If’,)) is positi\c. Moreover. dcspitc the fact that sullicient
conditions for the t\\o-par:lmctcr (mean-variance) modei arc not required for
(I 9, noncthclc>s IE(r,)-rfl varies directly with Std rj [provided u(r,. -_ V’
(IV,)) # 01. By multiplying (IS) by any portfolio proportion 11, and then
summing over j, it is also easy to SW that (15) must hold for any portfolio of
risky sccuritics. including the market portfolio of risky sccuritieh as a special
case.

“E is an expectation operator. Std a standard deviation operator, K the correlation co-

cflicicnt of r,. and - Y’( I+‘,.), and L is a constant as dctined and is nor IO be confused with a
Lagrangian multiplier. In the proof below. Cov is a covariancc operator. Hcja (1971) has
derived a valuation equation similar IO ( 15).except that il is not clearly linked to the construc-
tion of a composite individual. Morcovcr, in place of - V’(I{‘,). Ucja inserfs an undefined
‘market factor’. In this case. as Stephen Ross has mentioned to mc. Bcja’s valuation equation
is a tautology since then: must exist some random variable in place of - V’( W,) so that (1S)
is true for all j.
236 M. Rubinsrein. AA,gregarion rlteorem for securities markers

Generalizing from the discussion in Rubinstein (1973a. pp. 65,‘6), when no

risk-fret security exists,’ (15) will generalize to

E(rj) = E(rJ + k(rj, - V’( W,))Std rj, (ail il.

where jj is any portfolio with zero non-diversifiable risk; that is, ~(r~, - V’
(IV,)) = 0.
The popular two-parameter valuation equation is a special case of (IS).

Corollary (Two-parameter valuation). Whenwer a composite inciit~ithtalcan

bL>cotrstrtrctcd. i/ eitfw B = - I (quadratic u!ilit_r,) or the composi;c indiridual
makes twrttlal probahilil>~ asscssttwnts, then in equilibriutn

E(r,) = rF + l.K(rj, r,)Std r,, (all j). (16)

where At is the market portfolio of all securities and

- E[ I’“( Ij.,)]Std H’,

II=
E[ V’( lb.,)] .

Pm/: If B = -I, then Y’(W,,) = A- IV,,. Defining the rate of return

on the market portfolio of all sccuritics rMc zz lV,J( 11’”-CO), then I”( lP’,c) =
A - ( LV, - C,,)~&k. flq. (I(,) follows upon substitution of this into (IS). On the
other hand, if the compo$itc individual m;~kcs normal probability assc>~mcnts,
then as shown in Rubinytcin (1073~. pp. 613,‘4) Cov (rj, - V’( IV,)) = -E
[V”(IV,)] Cov (rj. IV,). Eq. (16) follows from this. Q.E.D.

Marc generally. Sul>iWsc sullicicnt coriditions to construct a composifc

individual arc not satisfied but all inilividu;~ls mahc homogcncou\ normal
probability asscssmcnts. As shown in Rubinstcin (1973c), valu;1tion equation
(16) will still hold cxccpt wilh the rcdcfinition

IioHever, in this ~215~2i. is stated in terms of individual-spccilic choice informa-

tion and the aggreg;ltion problem remains unsolved. Although this dots not
prevent (16) from yicldillg rcfut:lble cmpiric:ll hypothcscs as in Jensen (1972).
they arc not as strong as they would othcr\+ise be had the aggregation problem
been solved. \L’ithout its solution, little can be said about the size and tlctcr-
minants of A, except that I > 0. This contrast!, with the comparative static\
results available in Rubinstcin (1973~) whcrl the aggregation problem can bc
solved.

y\Vhcn no risk-fret wcurily cxia. WC mu31 cheek that n composiw individual still exis(s
smcc rhc proof in ;Iptxndi\ I relics on the cxistsncc of a rish-free sccurily. It csn bc shown
that a composi~s indlviclu:~t aids. s\cn in lhe ab~cncc of a risk-free sccuricy. in caws (i) and
(iv) and in caw (11) provided A = 0. Sw appenJi< 2.
 M. Rubinstrin. Aggregation theorem /or securities markers 237

When special restrictions, other than sufficient conditions to construct a

composite individual, are not placed on beliefs and tastes, the two-parameter
case can be generalized to a multiparameter valuation equation.

Corollary (Multiparameter valuation). Whenever a somposite in&dual can

be constructed, in equilibrium

(all i), (17)

w,frere rll is the market porlfolio of all scarrities,

- V’“‘( cv, - Co)“_ ’

0, = (all II 2 2).
(n- l)!E[C”( W,)] l

a.(rj,r4,) z E[(rj - Erj)(r,, - ErM)“- ‘1. (all j and II).

Proof. With the prior construction 0: a composite individual, the proof

follows immediately from Rubinstein (1973a. pp. 62-64). Q.E.D.

lntcrpreting valuation equation (17). the equilibrium expected rntc of return

ofany security equals the risk-fret rate plus a risk premium equal to the weighted
sum of the joint moments {(T,} of the rate of return of the security with the
market rate of return, whcrc each joint moment is weighted by an appropriately
normnlizcd dcrivativc of V cv:~lu;lkd at I:( IV,). Eq. (17) indicates that the
joint moments arc the appropriate itwu.wri*d of scwrily rid bcc;luw they rcflcct
the contribution of :I marginal incrcasc in the holdings of ;I security to the
corrchpontling central moments of future composite wealth. which arc the
:ippropri;ttc nwaswe.~ f/ porlffolio ri.vk in ;I niultiparamctcr model. Each joint
moment is wcightcd by tile ratio 0, rcllccting the corresponding mcdx~~ meawre
ofrisk aversion.
In brief, all discrctc time valuation models developed in the literature and
for which the aggregation problem has been solved arc special casts of (I 5).
Valuation equation (16) holds with composite quadratic utility or composite
normal probability assessments, and (17) is mcrcly an alternative form of (I 5).

5. Exchange-cfliciency

Exchange-efficiency is defined as circumstances under which individual5 arc

not motivated to crcatc exchange arrangcmcnts outside the market. Since the
economies satisfying the aggregation thcorcm of section 3 reach the same
allocation of present consumption and future wealth as would have cmcrged
with a complctc market, and since a complctc market is always cxchange-
effZcnt, then all economics satisfying the aggregation theorem must also be
exchange-cfflcicnt.
238 hf. Rubinstrin, AfgreRulion theorem for securities markets

A related theorem” will now be proven:

77reorem (Exchange-efficiency). Whenecer a composite indicidual can be

constructed, in equilibrium, the present l.alue of the sum of the returns on a set
of securities is equal 10 the swn of the present l:aIuec of the returns on those sccu-
rities.

Proqf. Define the aggregate present value of any risky security Sj, =
.S,l,/(l +rjr). Substituting this into (12) and (13)

s, = PE[y'(cvt)sj~l
JO
(all j).
WC,) ’
Consider any two risky securities j and k with summed returns or future values
sjI+skt- Clear!y, since

PEIV'(Wt)(Sjl +skl)l= PEIY’( rv,)Sj,] + PE[V’( tY*)S,*]

WC,) * U’(Co) WC,) ’
the present value of {.Si,.+S,,.} equals Sjo+Slo. This analysis is easily
extended to combinations of any number of securities. Q.E.D.

As one would cxpcct, whcncvcr a composite individual can be constructed

the packaging of sccuritics makes no difTcrcncc, and. in particular, corporate
capital structure (even if bankruptcy is possible) and non-synergistic mergers
(cvcn if bankruptcy is possible) arc irrclcvnnt.

6. Production-cfkicncy

Production-cfiicicncy is dcfincd as circumstances under which vnluc-maxi-

miring firms make Pareto-optimal production decisions. Since the economies
satisfying the aggregation thcorcm of section 3 rcnch the same allocation of
present consumption and future wealth as would hnvc emcrgcd with a complete
market, and since under appropriate compclitivc conditions a complctc market
is always production-cficicnt, then all economies satisfying the aggregation
theorem and the competitive conditions must also be production-cfflcicnt.
A rclatcd theorem will now bc proven:

‘I’heorcvn (Production-cficicncy). IVhen~wr a composite itwididuaf can bc

constructed. in equilibrium. competitiue r.allrc-l?lasirni;ir?r: firms make the same
production decirions us would the composite intliridual.

‘%ec Stiglitr (1969. p. 7901 for a verbal discussion of this thcorcm. and see Mocsin (1973.
p. 87) for a similar. although incorrect, version. One c;lvcclt is in order. When individuals
may have dilTcrcnt bclicfs [GIXS (v) and (vi)], wcr must check that eliminating the option ol
purchasing IWO securities scp;tr;ltely dots not undermind the complete markets condition
for tbc constructIon of 3 composite individual. See appendix 2.
 M. Rubinstein, Aggregation theorem fir securities markets 239

Proof. For any firmj letfj denote its production function mapping present
investment H,, and state e into future value Sjlc SO that S,tc = fj(Hjo, e)
where f; > 0 and f; < 0. This embodies the competitive assumptions of no
production externalities since fj is assumed independent of the production
decisions of other firms and decreasing returns to scale since fj is strictly in-
creasing and concave in Hi,. From the previous section, in equilibrium, the
present value Sj, of firmj is determined such that

S, = PEIY'(W*)Sjll = PEIY’(WIIfi(HjO)l
Jo
U’(Co) WC,) *
A value-maximizing firm will increase Hjo until dSjo/dHjo = 1 or alternatively
until
PE[U W,)f;wfo)l = ,
WC,)
This embodies the assumption that firm j is competitive in the sense that it
acts as if it cannof affect the aggregate amounts of production (C,“, {WE)) both
over dates and across states. Since f, is strictly concave, eq. (18) determines a
unique optimal level of investment Hi*, .
If instead of delegating production decisions to competitive value-maximizing
firms, the composite individual were to select the optimal investment levels
{H,‘,}, then he would solve the following programming problem:

max (/(Co)+/3 1 X.V(CY,.) S.t. Wo = Co+ 2 HjO/l,

co. (IlJOl c j
whcrc IV,, = ~,/,(rfJo, e)/I. Substituting in the constraint the problem becomes

Differentiating partially by /f,, the necessary and sufficient condition for an

optimum (given Co) is

-LI’(C,)+pE[Y’(W,)/;(llj+,)] = 0.

However, this condition, which uniquely determines If;,, is identical to eq. (18).
Consequently, a composite individual will make the same production decisions
as competitive value-maximizing firms.’ ’ Q.E.D.

Other papers - Fama (1972) and Stiglitz (1972) -have floundered on an

inappropriate definition of competition among firms. The definition used here,
which is consistent with Merton and Subrahmanyam (1973), requires absence

’ ‘This thcorcm holds even if each firm produces many commodilics. Simply redefine the
production function as S,,. = /,Cif,,' , il,,,‘, . . .. I/,o’. . . , ti,loc; e), where C(c = 1, 2.. . ..C)
is the number of commodities produced by firm j, and ditrerentiate partially with respect to
(H,,,‘). Observe it is unncccssary IO rcquirc that a*~(.)/a/f,,cil//,,C’ = 0. where c and c’ are
dilTcrcnt commodifics.
240 M. Rubinsrrin, Aggregation theorem for securities markers

of production externalities, decreasing returns to scale, and that firms act as

if they cannot influence the aggregate amounts of production over dates and
across states.

Appendix 1

Proof of Aggregation Theorem

(i) Proved in the text.
(ii) In light of section 2.3, rates of return of actual securities are determined
as if they were set in a complete securities market. Consequently, in place of
conditions (l)-(6), we can work instead with @)-(I I). When B # 0. U;(CA) =
(A,+ BCk)-b and V,‘(F’:.) = (Ai+ BWi,)-*, so that from (8)

(Ai+BC~)-* = (I +r,)pn,(Ai+ BWf,)-*,

therefore
Ai+ BC; = [(I +r,)prr,]-B(Ai+ BWf,).
Summing over all individuals

Using closure conditions (10) and (I I) and dividing by f,

(a) (A +X0)-* = (I +r,)pn,(A + BW,,)-*, (all e),

where Co z C,“/f, W,, = W,“,/l all c. :ind A = CiA,/l. Also using closure
conditions (IO) and (I I), from (9)

(b) IV, = C,+ C W,,(I+r,)-‘,

c
where W,, I W,“// E ~,W~/l. Observe that (a) and (b) are the necessary
and sufficient conditions for an optimum that would have been derived had
there existed an individual with the specified composite resources, beliefs, and
tastes. Since rates of return arc determined as if thcrc were a complete market,
conditions (l)-(3) applying to the actual economy must hold for a composite
individual.
(iii) Again rates of return on actual securities are dctermincd as if there were
a complete market. The proof follows immediately from the proof for (v).
(iv) In light of section 2.3, rates of return of actual securities are determined
as if they were set in a complete market. Consequently, WC can work directly
with conditions @)--(I I). When A = 0 and B = I, c/{(C$ = (CA)-’ and
V,‘(Wf,) = (Wit)- I, so that from (8)
W:, = (I +r,)p,qCA .
Summing this over e and using (9). W,, - CA = piCA so that Cb = W,( I +p,)- ‘.
 hf. Rubinstein. Aggregation theorem for securities markers 241

Substituting this into the above equation and summing over all individuals

C wf, = (l+r,)n,W,Cpi(l+pi)-L,
I i

using closure condition (11) and dividing by 1.

w,, = (1 +r,)J%W,[ 1 P,(l +P,)- ‘Irl.

i
where W,, E WE/f. By defining composite rate of patience,

P z ~‘-_(~(~+Pi)-‘l~}/[~(i+Pi)-‘/‘l~
‘ i

then &i(l+pi)-L/I = p(l+p)-* so that

IY,, = (l+r,)pn,W,(l+p)-‘.

Since CA = CV,(I +pJ-’ and using closure condition (IO), C,, = W, xi

(I+pJ-‘Y where C, = C,“/f. From the above definition of p, (I +p)- ’ =
xi<1 +pJ-‘/J, so that C, = W,(I i-y)-‘, and

(a’) IV,, = (1 +r,)pn,C,. (all e).

Also using closure conditions (IO) and (I I), from (9)

(b’) rv, = c,+ c IY,,(I +r,)_‘.

(
From this point the argument is identical to (ii).

(v) When B = 0, U,f(Ch) = cxp (-CL/A,) and I’;( Wf,) = cxp (- lY~,/A,)
so that from (8)

exp (- C&f,) = (I + r,),‘p; cxp (- Iv;,/.4 J,

therefore

-CA = A,ln(l +r,)+Ailn,J,+Ailnn~- IV:,.

Summing over all individuals

- 7 CA = $ Ailn(l +r,)+lr.(lj~~~“‘)+In(~ nL”)- q CVY,.

Using closure conditions (IO) and (1 I) and dividing by f,

-Co = Aln(l +r,)+In(v lJiA”‘)+In(fl rryd”)- IV,,,

i
where C, = Ct/I, W,, = WfA,/I and A E xiAi//. Dividing by A,

-Co/A = In(l +r,)+In p+ln x,- W,,/A,

where p z nipiA6’E’A’ and rr< 3 n,~~:“*“~.‘l. Consequently,

(a’) exP (-G/A) = (1 +r,)pn, exp (- bt’,,/A), (all e).

242 M. Rubinstein. Aggregafion theorem fir securities markets

Also using closure conditions (10) and (1 I), from (9)

(b”) w, = c()+ 1 Wle(l+fc)-‘,

c
where IV,, = K’bw,‘f.Observe that (a”) and (b”) are the necessary and sufficient
conditions for an optimum that would have been derived had there existed an
individual with the specified composite resources, beliefs, and tastes.
(vi) When A = 0 and B = I, from (8)

cv:, = (I +r,)pn;C;.

From the proof of (iv), since CA = W,(I +p)- ’ by closure condition (IO),

w:, = (I +rc)P&Or
where C, E C’i’/I. Summing over all individuals, using closure condition (1 I)
and dividing by I,

(a? w,, = (1 +r.)PJ&r

where IV,, E WE/f and 7~~E xi~i/f, all CJ. Also using closure conditions
(IO) and (I I), from (9)

(b”) IV, = co+ 1 W,,(l +rc)-‘.

I?
From this point the argument is identical to (v). Q.E.D.

Appendix 2

Op!itnal siiaring rulcas

Many of the results of this paper arc better understood from the pcrspcctive
of optimal shurin~ ruk~s. With the homogcncity Conditions of cast (ii), it can
be shown that

Iv/’ = AWll-niwO .-/,a+SW;,

Ir IV It* (all E and i),
ACT+ SW, + kj+BW,

where 4 = I +(I -~-r~)-l. The last term in the summation has the natural
intcrprctation of a state-dependent tlici&nt/ and the second term a certain
side paytncwt. Observe that

~(A~v~-ni~v~)(A&+O~Y,)-’ = 0,

; (Aid, + BbVA)(/I(/J + BCV”)_ ’ = 1.

Conscqucntly, despite the variety of securities nvail:lb!c, perhaps a complete

market, every individual chooses to hold the m;lrkct portfolio M in proportion
 M. Rubinstein. Aggregation theorem for securities markets 243

(Ai~+BW~)(A~+B~V,,)-’ and to borrow or lend (AW~-AilY,)(A~+BW,)

(1 +rF)- ’ at certain rate of return rF (universal portfolio separation). When
composite parameter A = 0. the sharing rule reduces to Wf, = (WA/W,) Wlrr
and no side payments are made.
With the homogeneity conditions of cases (iii) or (v) (exponentizl utility),
the sharing rule can be shown to be

Fvf, = ~,ln(nf/lr.)+.-l,[ln(p,/p)- C (1 +re)-‘In(nf/rr,)]$-’

c

+(A ~v~-Ai~~~)(A~)- ’ +(Ai/A) Wlc, (all e and i).

The last term is a dividend, the second the third terms a side payment, and the
first term may be interpreted as a side bet. Observe that

7 Ailn(I$n,) = 0,

T A ,[ln(pJp) - 5 (I +rc)- ‘ln(&n,)]f$- ’ = 0.

c (A WA--A,tV,)(A~)- ’ = 0,
‘

When all individuals have the same bclicfs for all states e. then no side bets
arc made and thcrc is universal portfolio separation [case (iii)]. Moreover,
the USC of side bets only arises for those states c for which IK: # n, for some
individuals i. Consequently, a complctc market is not strictly required in case
(v). For those states c for which rr: = X, for all individuals i, state contingent
cl;tims need not exist.
Kubinstcin (1973d, p. 25) suggests that speculative volume can be used as
an index of the ‘information-cliicicncy’ of the securities market (i.e., the extent
to which security prices fully rcllcct information). With cxponentisl utility,
as has been shown, spcculativc volume is mcasurcd by

7 ~f,M~!./d- 4-’ 1
C
( I +fr)- ‘In(nf./%,)].

If bclicfs are the same for all individuals for all statrs, then thcrc will bc no
spcculativc trading, independent of tastes as measured by {Ai}. However,
given the decision to speculate, the volume of speculative trading does depend
on (Ai}.
With the homogcncity conditions of casts (iv) and (VI) (logarithmic utility),
Arrow (1972) has shown that the optimal portfolio proportions are detcrmincd
by the surprisingly simple decision rule /I: = ~1 for all e and i. Clearly, to the
cxtcnt individuals agree on the probability of sonle states. the securities market
may bc incomplctc.
244 M. Rubinsfein, Agpregafion rheorem for securities markets

References
Arrow, K.J., 1972. The value and demand for information, in: C.B. McGuire and R. Radner,
eds.. Decision and organization (North-Holland. Amsterdam) 131-140.
Baja, A., 1971. The structure of the cost of capital under uncertainty, Review of Economic
Studies, 359-368.
Fama, E.F., 1972. Perfect competition and optimal production decisions under uncertainty,
Bell Journal of Economics and Management Science, 509-530.
Hakansson, N.H., 1970. An induced theory of the firm under risk: The pure mutual fund,
Journal of Financial and Quantitative Analysis 5, 155-l 78.
Jensen, M.C.. 1972, Capital markets: Theory and evidence, Bell Journal of Economics and
Management Science 4, 357-398.
Lintner, J.. 1969. The aggregation of investor’s diverse judgments and preferences in purely
competitive markets, Journal of Financial and Quantitative Analysis 4. 346-382.
Lintner, J.. 1972, Security market equilibrium, price of risk, and size of market: A generaliza-
tion, Harvard Institute of Economic Research Report no. 234.
Merton, R. and M. Subrahmanyam, 1974. The optimality of a competitive stock market,
Bell Journal,of Economics and hlanagement Science.
Mossin. J., 1973. Theory of financial markct5 (Prentice-Hall. New York).
Pye, G., 1967. Portfolio selection and security pricec, Review of Economics and Statistics
49,111-115.
Rubinstein. M.. 1973a. The fundamental theorem of paramctcr-preference security valuation,
Journal of Financial and Quantitative Analysis 8, 61-69.
Rubinstcin. M., 1973b. A mean-variance synthesis of corporate financial theory, Journal of
Finance 28. 167-182.
Rubinstein, M.. 1973~. A comparative statics analysis of risk prcmiumr, Journal of Business
46, GOS-615.
Rubinstcin. M., 1973d. Securities market etlicicncy in an Arrow-Dcbrcu cu-onomy. Rcscarch
Program in E‘inancc Working Paper no. 14 (Bcrkclcy, Calif.).
Stiglitr, J.E.. 1969, A rc-examination of the Modigliani -Miller thcorcm, American Economic
Review 59. 784-793.
Stiglitr. J.E.. 1972, On the optimality of the stock market allocation of invcctmcnt. Quarterly
Journal of Economics 86, 25 -00.
Wilson, R., 1967. A Pareto-optimal dividend policy, Managcmcnt Scicncc 13. 756-764.
Wilson, R., 1968, The theory of syndicates, liconomctricn 36. IIY-132.