Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Technical Memorandum Number 724F
Discounting and Risk Neutrality
by
Matthew J. Sobel
Revised September 2006
Department of Operations Weatherhead School of Management
Case Western Reserve University 330 Peter B Lewis Building
Cleveland, Ohio 44106
DISCOUNTING AND RISK NEUTRALITY‡
Matthew J. Sobel**
September 1990; Revision September 15, 2006
Abstract
Let be a binary relation in which is a real vector space of vector-valued discrete-timeÐ ¤ ß Z Ñ Z
stochastic processes with as the sequence of zero vectors. If satisfies four axioms, then! Ð ¤ ß Z Ñ
there are unique discount factors such that preferences regarding stochastic processes induce prefences
among present value random vectors. Three of the axioms are familiar: weak ordering, continuity, and
non-triviality. The fourth axiom, , is implies where isdecomposition \ ] ¤ Ð £ Ñ \ ¤ Ð £ Ñ ]! !
the sequence of zero vectors. Also, if preferences satisfy the four axioms then the following properties
are equivalent: the converse of decomposition, the existence of a felicity function on the set of random
vectors, and risk neutrality. In this limited sense, discounting implies risk neutrality.
JEL
D81, D99, G12
*In honor of Martin Shubik and in memory of Robert Rosenthal.**Department of Operations, Weatherhead School of Management, Case Western Reserve University,Cleveland, OH 44106-7235.
Samuelson (1937) proposed the discounted utility model in a deterministic framework which was
subsequently axiomatized by Koopmans (1960), Williams and Nasser (1966), and others. An
axiomatic foundation for discounting in a stochastic setting, the subject of this paper, addresses the
basic properties of intertemporal preferences which yield utility functions having particular structures.
Since the discounted utility model is additively separable, the literature on separability is relevant to
this issue (Blackorby, Primont, and Russell (1998)). In spite of this considerable attention,
inconsistencies abound in professional practice and in research literatures.
Under a strong form of Koopmans' axioms, there exists an utility function and aintraperiod ?
discount factor such that, for all deterministic scalar sequences ( , , ...) and ( ," − Ð!ß "Ñ B œ B B C œ C" # "
C#, ...),
(1) is weakly preferred to ( ) ( )B C Í ? B ? C! !> >
>" >"> >" "
If and were replaced by stochastic processes ( , , ...) and ( , , ...), then each sideB C \ œ \ \ ] œ ] ]" # " #
of the inequality would be a random variable. Would a combination of Koopmans' axioms and those
from utility theory von Neumann and Morgenstern (1944) for example, imply the existence of anß
interperiod utility function such that@
(2) is weakly preferred to [ ( )] [ )]\ ] Í I @ ? \ I @ ?Ð]Š ‹ Š ‹! !> >
>" >"> >" "
(where denotes expected value)?I
In the axiomatization of preferences among scalar sequences by Williams and Nassar (1966), is?
the identity function. So
(3) is weakly preferred to B C Í B C! !> >
>" >"> >" "
Would a combination of axioms from Williams and Nassar and utility theory lead to
(4) is weakly preferred to [ ( )] [ ( )]\ ] Í I @ \ I @ ]! !> >
>" >"> >" "
The preference model in numerous research fields is
(5) is weakly preferred to ) )\ ] Í IÒ ?Ð\ Ó IÒ ?Ð] Ó! !> >
>" >"> > " "
2
for an intraperiod utility function . Of course this is (2) with risk neutrality and risk sensitivity,?
respectively, concerning interperiod utility and intraperiod utility. An elementary ordering that is
widely used in professional practice and several research fields is
(6) is weakly preferred to ) )\ ] Í IÐ \ IÐ ]! !> >
>" >"> > " "
Of course this is (2) with risk-neutral interperiod and intraperiod utility functions, and it is (5) with a
risk-neutral intraperiod utility function.
Alternative decision rules are compared via (5) and (6) in diverse areas such as accounting
[Moriarity and Allen (1984)], advertising [Monahan (1983)], banking [Shubik and Sobel (1992)],
biology [McNamara (1990), Mendelssohn (1979)], capital accumulation [Sethi (1998)], energy
[Murphy, Toman and Weiss (1989)], engineering [Yeh (1985)], finance [Altug and Labadie (1994),
Cochrane (2000), Duffie (2001), Sharpe (1985), Ho and Lee (1986)], game theory [Shapley (1953)],
production [Arrow, Karlin and Scarf (1962)], replacement [Jorgenson, McCall and Radner (1967)],
social psychology [Shubik (1970)], and technological change [Balcer and Lippman (1984)].
Can (2), (4), (5), and (6) be reconciled? Do successively stronger assumptions underlie each of (2),
(4), and (6) and (2), (5), and (6)? Researchers utilize (5) instead of (6) to investigate the effects of risk
sensitivity. So it is not interesting that (2), (4), (5), and (6) are equivalent if the decision maker is risk
neutral.
There is a fundamental difficulty with using (5) to investigate the effects of risk sensitivity. When
axioms that imply the existence of discount factors (Theorem 1 in §2) are augmented with the
existence of an period utility function, then that function is linear (Theorem 2 in §3). Professorintra
James E. Smith, Duke University, notes that (4) implies that an interperiod utility function is an
intraperiod utility function in a single period mdel. So it too must be linear. In this limited sense,
discounting implies risk neutrality.
Miyamoto and Wakker (1996) unify much of the literature on multiattribute preference orderings
and utility functions that is reviewed by Dyer and Sarin (1979), Farquhar (1977), Fishburn (1978), and
Keeney and Raiffa (1976). Wakker (1993) has axioms implying that there are positive numbers ""ß
"#ß ÞÞÞ <Ð † Ñ \ ]and a real-valued function of random variables such that is weakly preferred to if and
3
only if where . Let be a degenerate random variable that takesYÐ\Ñ YÐ] Ñ YÐ\Ñ œ <Ð\ Ñ HD "> > > +
the value with probability one, and let be the real-valued function on the reals such that+ A
AÐ Ñ œ <ÐH Ñ <a . The linearity of (in probabilities in the sense of (11) in §3) implies+
Y Ð\Ñ œ IÒ AÐ\ ÑÓÞ AÐ † ÑD "> > > This paper shows that is linear ( , risk neutral) under weak sufficienti.e.
conditions for additive separability. This result diminishes the justification for (5).
Literatures which depend on (5), such as asset pricing and, more generally, theoretical financial
economics, are left in an ambiguous position. However, individuals exhibit sensitivity to riskdo
(gambling, insurance, etc.); so this paper casts further doubt on the descriptive validity of conclusions
predicated on expected present value. Experimental work challenges expected utility theory and
comparisons of deterministic temporal sequences via present values [Frederick, Loewenstein, and
O'Donoghue (2003)].
A relatively small body of work provides axiomatic justifications for comparing deterministic
sequences via their present values. Koopmans (1960) and Koopmans, Diamond, and Williamson
(1964) study orderings of countable sequences of vectors of real numbers. They postulate the
existence of an interperiod utility function such that deterministic sequence is@Ð † Ñ B œ B >
weakly preferred to if and only if ( ) ( ). They obtain sufficient conditions for theC œ C @ B @ C>
existence of ( ) and an intraperiod utility function such that" ‘ ‘− !ß " ? À ÄQ
@ B œ ?ÐB Ñ( ) . Koopmans (1972) specifies conditions on a binary relation which imply the!_>œ"
>">"
existence of with the properties assumed in the earlier studies.@Ð † Ñ
Williams and Nassar (1966) obtain sufficient conditions for a weak ordering on a real vector_¤
space of finite sequences ( ,..., ) of real numbers to have the following property: there existsB œ B B" X
" "− " B ¤ C B C (0, ) such that if and only if ( ) 0._ !X>œ"
>"> >
Lancaster (1963), Fishburn (1970), and others cited by Fishburn and Rubinstein (1982) and
Frederick, Loewenstein, and O'Donogue (2003) investigate closely related deterministic models of time
preference and impatience. Epstein (1983) studies a stochastic model that relates time preference and
risk preference. He specifies necessary and sufficient conditions for a binary relation on the set of_¤
real-valued discrete-time stochastic processes to have the following property. Let ( , ,...) andB œ B B" #
4
C œ C C( , ,...) be degenerate ( constant with probability one) real-valued stochastic processes." # i.e.,
There exists (0, ) and such that if and only if _" ‘ ‘ "− " ? À Ä B ¤ C ?ÐB Ñ!_>"
>">
?ÐC Ñ ?!_>"
>">" . Technical differences aside, this paper presents conditions under which is affine.
The axioms in this paper are briefly compared in §1 with the assumptions made by Williams and
Nassar, Koopmans, Epstein, and Miyamoto and Wakker.
Meyer (1976) uses the same rationale that leads from (1) to (2). He develops a cardinal comparison
of alternative sample paths which are outcomes in the sample space, say of ( , , ...) andB œ B B" #
C œ C C( , , ...), and then, observing that the cardinal measures are random variables, he applies" #
expected utility theory. He implicitly assumes orthogonality of the axioms underlying time preference
and risk preference. The present paper answers a question addressed to the author by Robert
Rosenthal (1987) who wondered if the two axiom systems might be related. They are indeed related.
It is not clear how the previously cited studies and this paper are are related to investigations of
consistency in intertemporal choice preference structures. See Kreps and Porteus (1979), Johnson and
Donaldson (1985), Machina (1989), and their references.
Section 1 presents notation and discusses axioms. Section 2 has sufficient conditions for
discounting, , four axioms that imply that comparisons between vector-valued stochastic processesi.e.
correspond to comparisons between their present values (which are random vectors) A further axiomÞ
yields a vector of familiar geometric discount factors. Three of the axioms are familiar: weak ordering,
continuity, and non-triviality. Instead of an independence or Archimedean axiom, the fourth is
decomposition, namely implies where is the sequence of zero\ ] ¤ Ð £ Ñ \ ¤ Ð £ Ñ ]! !
vectors. Section 3 shows that if preferences satisfy the four axioms then the following properties are
equivalent: the converse of decomposition, the existence of a felicity function, and risk neutrality.
In §4, a combination of results in earlier sections with some in Koopmans (1972) yields sufficient
conditions for a scalar discount factor to generate vectors of geometric discount factors. Section 5
briefly examines relationships among alternative continuity axioms and definitions of impatience.
Section 6 sketches the implications of §3 for attribute decomposition in random vectors.
5
1. Notation and Axioms
Fix a probability space ( , ,P). Let and be respective sets of positive integers and countableH Y M Z
stochastic sequences ,... defined on ( , ,P) with for all .\ œ Ð\ ß\ Ñ \ Ð Ñ − Ð>ß Ñ − M ‚" # >QH Y = ‘ = H
Let denote the - component of . If , , and , let and\ 7 \ \ − Z ] − Z , − \ ] − Z7> >th ‘
,\ − Z Ð>ß Ñ − M ‚ \ ] ,\ have values at given by ( ) ( ) and ( ). Let be the zero vector in= H = = = )> > >
‘ ) )Q , and let ( , ,...) . So is a real vector space with zero element . ! !œ − Z Z
Let be the - unit vector in and let be the sequence of vectors except that the/ 7 / − Z7 7>Qth ‘ ) !
t- vector is . For ( ) and ( ) let denote .th / ? œ ? − œ − ? † ?7 3 3 3 3Q Q Q
3œ"‘ 8 8 ‘ 8 8! Let be a binary relation on , write if and , and write if ¤ Z \ µ ] \ ¤ ] ] ¤ \ \ ¢ ] \ ¤ ]
but not . The following notations are written interchangeably: and , and] ¤ \ \ ¤ ] ] £ \ \ µ ]
] µ \ \ ¢ ] ] ¡ \, and and . A (sometimes called a ) is a completeweak ordering strong ordering
transitive binary relation.
Let be the set of -valued random vectors. For , let ( , ) denote ( , , ,...). ThenW G − W G − Z G‘ ) )Q !
( , ) induces the following binary relation on : if and only if ( , ) ( , ). Let_ _¤ Z ¦ W E ¦ F E ¤ F! !
[ ¸ ^ [ ¦ ^ ^ ¦ [ [ ¦ ^ [ ¦ ^ ^ ¦ [ denote and and let denote but not . Notations_ _ _ _
employed interchangeably include and , and , and and_ _[ ¦ ^ ^ ¥ [ [ ¸ ^ ^ ¸ [ [ ¦ ^
^ ¥ [ E − W IÐEÑ −. For , let denote the vector of expectations of components (when the‘Q
expectations exist).
Axioms
In the following axioms, , , and\ œ Ð\ ß\ ß ÞÞÞÑ − Z ] œ Ð] ß ] ß ÞÞÞÑ − Z ß > − 7 œ "ß ÞÞÞßQ" # " # Mß
- 0.
(A1) Transitivity and rationality: weakly orders¤ Z
(A2) Decomposition: implies\ ] ¤ Ð £ Ñ \ ¤ Ð £ Ñ ]!
(A3) { : } is closedContinuity: ! !\ ] ¤ Ð £ Ñ !
(A4) More is better: / ¢7> !
(A5) ( , ,...) ( , , ,...) Stationarity: implies\ \ µ \ \ µ" # " #! !)
(A6) ( )Sooner is better: \ / / ¢ \- 7> 7ß>"
6
The first four axioms lead to a present value formula. The fifth axiom yields the usual geometric
form of discount factors; the sixth axiom implies that discount factors are less than one. A loose
interpretation of (A5) is that indifference between receiving a cash flow and not receiving it implies
indifference for the one-period delay.
Reasonable alternative versions of the continuity axiom (A3) include the following statements:
(A3 ) 0 0w \ ¢ Ð ¡ Ñ \ ¡ Ð ¢ Ñ\ ! implies that there exists such that if! ! ! !* *
(A3 ) ww \ ¤ Ð £ Ñ Ê ,\ ¤ Ð £ Ñ a , !! !
However, (A3 ) and (A3 ) are equivalent to each other and to the restriction of (A3) in which w ww ] œ !
(Proposition 5 in §7).
Decomposition axiom and its converse
Axiom (A2) is particularly important and may be contrasted with its converse:
(A2 ) ( ; )c \ ¤ Ð £ Ñ ] Ê \ ] ¤ Ð £ Ñ \ − Z ] − Z!
The present value formula in §2 invokes (A2) but not (A2 ).-
Proposition 1: Suppose (A1) is valid.
(a) (A2) [ , , ]Í [ ¤ Ð £ µ Ñ Ê [ ^ ¤ Ð £ µ Ñ^ a ^ − Z!
(b) (A2 ) [ , , ]c Í [ ^ ¤ Ð £ µ Ñ^ Ê [ ¤ Ð £ µ Ñ !
Proof: For sufficiency in (a), use and . For necessity in (a), use and\ œ [ ^ ] œ ^ [ œ \ ]
^ œ ] \ œ [ ^ ] œ ^. For (b), let and .
Thus, with (A2), is as good as the status quo only if incrementing with is as good as .[ ^ [ ^any
With (A2 ), is not as good as the status quo if there is which is better than augmented by- [ ^ ^any
[ . The combination of (A2) and (A2 ) is a version of the “independence" assumptions that recur (andc
whose descriptive validity is questioned) in axiomatic theories of decision making. Also, this
combination corresponds to “persistence," “monotonicity," and “consistent choice" in dynamic
programming [Denardo (1967), Sobel (1975, 1980), and Blair (1984)]. Two applications of part (a)
imply
[ ] [ ]\ ] µ \ µ ] Í [ µ Ê [ ^ µ ^ a ^ − Z! !Ê
7
The counterpart to in Miyamoto and Wakker (1996) can be construed as a set of sequences ofZ
time-indexed marginal probability distributions. Allowing for difference in the models, their axioms
include (A1) and (A3) and, instead of (A2), assumptions of outcome monotonicity and the Thomsen
condition. Define outcome monotonicity here as and implyE ¤ Ð £ ÑF [ ¤ Ð £ Ñ^ E [ ¤
Ð £ ÑF ^ . The following result shows that (A2) alone is less restrictive than outcome monotonicity
because the former is equivalent to (A2) together with (A2 ).c
Proposition 2: If satisfies (A1) then it satisfies outcome monotonicity if and only if it satisfiesÐ ¤ ß Z Ñ
both (A2) and (A2 ).c
Proof: (A2) and (A2 ) outcome monotonicity:c Ê
Using (A2 ), and imply and So Lemma 1(b) in §2 impliesc E ¤ F [ ¤ ^ EF ¤ [ ^ ¤! !.
E[ F ^ ¤ E[ ¤ F ^! yielding by (A2).
:Outcome monotonicity (A2)Ê
Let , , and . Thus, and are and . SoE œ \ ] F œ [ œ ^ œ ] E ¤ F [ ¤ ^ \ ] ¤ ] ¤ ]! !
outcome monotonicity, , is .E[ ¤ F ^ \ ¤ ]
:Outcome monotonicity (A2 )Ê c
Let , , and . Thus, and are and . SoE œ \ F œ ] [ œ ^ œ ] E ¤ F [ ¤ ^ \ ¤ ] ] ¤ ]
outcome monotonicity is . \ ] ¤ !
Preferences are if implies for every that isconstant risk averse \ ¤ ] \ ¤ ] − Z% % %
constant with probability one. However, and (A2 ) imply \ ¤ ] \ ] ¤ \ ¤ ] c ! Ê % %
from (A2). Preferences are if implies for all . However,relative risk averse \ ¤ ] +\ ¤ +] + !
\ ¤ ] \ ] ¤ Í +Ð\ ] Ñ ¤ + !, (A2 ), and Lemma 2(a) in §2 imply for all . Then (A2)c ! !
yields . Since axioms (A1), (A2), (A2 ), and (A3) imply both constant and relative risk+\ ¤ +] c
aversion, one should anticipate risk neutrality (cf. Theorems 1 and 2 in Miyamoto and Wakker).
Axiom sets
The first four axioms imply that there are discount factor vectors such that a vector-valued stochastic
process is weakly preferred to another if and only if there is a corresponding weak preference
concerning the present value random vectors. Conversely, axioms (A1) and (A3) are satisfied by any
8
preference relation among stochastic processes that is determined by preference among present value
random vectors. Also, (A4) is satisfied if the preference among random vectors has the property that
more is preferred to less. However, (A2) is necessarily satisfied only if and are independent. So\ ]
the key axiom is decomposition, namely (A2).
The following example satisfies (A1), (A3), and (A4) but neither (A2) nor (A2 . Let ,-Ñ Q œ X œ "
let denote preferences among random variables corresponding to variance minus expected value,¦
H œ Ö+ß ,× + , \Ð+Ñ œ !ß \Ð,Ñ œ "ß ] Ð+Ñ œ "ß with probabilities 3/4 for and 1/4 for , and and
] Ð,Ñ œ ! \ ] ¦ ] ¦ \ \ ] ¢ ] ¢ \. Then and , so and .) !
A loose interpretation of (A5) is that indifference between receiving a cash flow and not receiving it
implies the same indifference for the one-period delay. In §4, the converse of (A5) implies that a
scalar discount factor generates a vector-valued discounte factor.
Consider restrictions of the assumptions to finitely long deterministic sequences of scalars. The
axioms in Williams and Nasser (1966) correspond to (A1) through (A6) (with (A2 ) and a continuityc
assumption different from (A3)). The postulates in Koopmans (1972) roughly correspond to (A1)
through (A5), (A2 ) and the converse of (A5). It is surprising that “sooner is better" is implied by hisc
postulates (although he does not directly assume the counterpart of (A6)). The preference ordering in
Epstein (1983) is defined on the space of probability measures on the Borel field of countable
sequences of real numbers. Epstein's assumptions, a restatement of Koopmans' in a stochastic
framework, correspond roughly to (A1) through (A5), (A2 ), and the converse of (A5). The continuityc
assumptions in Williams and Nassar (1966), Koopmans (1972), and Epstein (1983) differ from each
other and from (A3).
2. Discounting Theorem
This section owes much to Williams and Nassar (1966). Its main results, Theorem 1 and
Corollaries 1 and 2, are stated and then proved with several lemmas.
It is convenient to define for each , ( ,..., ) for each , and for to write" " " "7" > "> Q>œ " 7 œ > \ − Z
" "> > 7> 7>\ 7 \ ¤ Z for the random vector with component . Recall that ( , ) induces the binaryth
9
relation ( , ) on the set of random -vectors. The following result gives sufficient conditions for_¦ W Q
preferences in finitely long processes to be consistent with a stochastic discounting formula. The proof
would be simplified by replacing the assumption of (A2) with the stronger assumption of outcome
monotonicity.
Theorem 1: (A1), (A2), (A3), (A4) 0 ( Axioms and imply the unique existence of for each"7> 7
and such that>)
(7) _\ ¤ ] Í \ ¦ ]! !X X>œ" >œ"> > > >" "
for all and with ,..., , ) ,..., , ) .\ œ Ð\ \ ] œ Ð] ] X − M" X " X0 0
The proof of Theorem 1 uses the property that , , implies ] µ 4 œ "ß ÞÞÞß N ] µ( ) ( )4 4N4œ"! !!
(Lemma 1(d)). The following countable analog is one of several ways to extend the theorem:
(8) If for all and then .] µ 4 ] − Z ] µ( ) ( ) ( )4 4 4_ _4œ" 4œ"! !! !
Corollary 1. (A1), (A2), (A3), (A4) (8)Axioms and and imply
_\ ¤ ] Í \ ¦ ]! !_ _>œ" >œ"> > > >" "
if and exist and have finite components with probability one .! !_ _>œ" >œ"> > > >" "\ ]
Now the stationarity axiom (A5) yields geometric discount factors.
Corollary 2. (A1), (A2), (A3), (A4), (A5) Axioms and imply for each there uniquely7 œ "ß ÞÞÞßQ
exists such that for all If also is valid then ( ) . (A6) ." " " "7 7> 7 7>" ! œ > − M "
Proof of Theorem 1
Lemma 1. Assuming (A1) and (A2):
(a) ( , ) ( , )\ ¢ ¤ µ Í \ ¡ £ µ! !
(b) , , ,\ ¤ Ð £ µ Ñ ] ¤ Ð £ µ Ñ Ê \ ] ¤ Ð £ µ Ñ! ! !and
(c) \ ¤ Ð £ Ñ ] ¢ Ð ¡ Ñ Ê \ ] ¢ Ð ¡ Ñ! ! !and
(d) \ µ ] µ Ê \ ] µ! ! !and
(e) \ ] µ Ê \ µ ]!
Proof. (a): due to (A2). So .\ £ Í \ £ Ê \ ¤ \ ¤ Í \ £! ! ! ! ! !
Now and and \ µ Í \ ¤ \ £ \ £ \ £ Í \ µ Þ! ! ! Í ! ! !
10
Now and not So not Also, So\ ¢ Í \ ¤ \ µ \ µ Þ \ ¤ \ £! ! ! ! ! Í !. .
\ ¡ Þ!
(b) and (c): The contrapositive of (c) is implies , , or both. From (A2),\ ] £ \ ¡ ] £! ! !
\ ] £ \ £ ] \ ¤ ] ¤ £ ] ¢! ! ! ! ! implies . So (A1) and imply and (a) yields Y . If
then and (a) imply . Using (c), (b) follows from [ ]\ £ ] \ ¡ \ ¤ ] µ Ê \ ] ¤! ! ! !and
due to Proposition 1(a) and (A1).
(d): Two applications of (b).
(e): Two applications of (A2).
Lemma 2. Assuming (A1), (A2), (A3):and
(a) 0.\ ¤ Ð £ Ñ Í -\ ¤ Ð £ Ñ - ! ! for all
(b) and .! £ Ð ¤ Ñ\ ! Ÿ - - Ê -\ £ Ð ¤ Ñ Ð- Ñ\# #
(c) .\ µ -\ µ - −! !Ê for all ‘
(d) 0 - Ê\ µ - Á \ µ Þ! !and
Proof. (a): is trivial ( ) and is trivial if . For when and (positiveÉ - œ " Ê - œ ! Ê \ ¤ - − M!
integers), Proposition 1(a) implies so . Inductively, and\ £ #\ £ #\ 8\ £ Ð8 "Ñ\!
! !£ Ð8 "Ñ\ -\ ¤ - − M - −. So . Let where is the set of positive rational numbers.for all g g
So with and , and . Therefore, if If - œ 7Î8 7 − M 8 − M - œ \Î8 -\ ¤ \Î8 ¤ \Î8 ¡!"
7
! !Þ !
then Proposition 1(a) with yields ; so (A1) implies [ œ ^ œ \Î8 \Î8 ¤ #\Î8 #\Î8 ¡ !.
Inductively, and for In particular, SoÐ5 "Ñ\Î8 £ 5\Î8 5\Î8 ¡ 5 − MÞ \ œ 8\Î8 ¡ Þ! !
\Î8 ¤ \ ¤ Þ -\ ¤ - −! ! !because Therefore, for all .g
If , let where ( ] and { }. Hence, and there exists a-  - œ 8 0 0 − !ß " 8 − M ! 8\ ¤g !
sequence with as . So for all , (A3), and Lemma 1(b) imply 0 − 0 Ä 0 3 Ä _ 0 \ ¤ 3 0\ ¤3 3 3g ! !
and -\ œ 8\ 0\ ¤ Þ!
(b): , (a), and Proposition 1(a).! ! !£ \ £ -\ £ \, , #
(c) [ and ] (by Lemma 1(a) and (a)) for all \ µ Í \ ¤ \ £ Í -\ ¤ - − d! ! ! !
(d) Use Lemma 1(a) and replace with in (a) replace with and with .- ,ß \ ,\ - ,"
11
The proof of Lemma 3(a) uses an argument due to Professor James C. Alexander, Department of
Mathematics, Case Western Reserve University.
Lemma 3. (A1), (A2), (A3) (A4):Assuming and
(a) For all and there uniquely exists such that ... 0 > œ #ß $ß 7 œ "ß ÞÞÞßQ "7>
.! µ / /7" 7>"7>"
(b) Also assuming (A6) implies ." "7ß>" 7> "
Proof. (a): If and with and thenUniqueness - ! - ! / -/ µ / - / µw w7" 7> 7" 7>! !
/ - / µ Ð- - Ñ/ µ7" 7> 7>w w! ! from Lemma 1(a), so from Lemma 1(b). Uniqueness follows from
(A4) and Lemma 2(d).
Let . If then for all large enoughExistence E œ Ö À / / ¢ × E œ g / Ð"Î Ñ/ £! ! !7> 7" 7> 7"! !
! (Lemma 2(a)), so (A3) implies . But so ./ £ / ¢ E Á g7> 7>! !
Let . If then so ((A2) and (A4)). Now,- œ 380Ö − E× − E / / ¢ / ¤ / ¢! ! ! !7> 7" 7> 7"! !
! is neither zero ((A1)) nor negative (Lemma 2(a) and (A4)). So .- !
From (A3), For any , so from (A3).-/ / ¤ Þ - / / £ -/ / £7> 7" 7> 7" 7> 7"! ! !! !
Therefore, . If then so (Lemma 1(a)); but ((A4)).-/ / µ - œ ! / µ / µ / ¢7> 7" 7" 7" 7"! ! ! !
So . Let .- ! œ "Î-"7>
(b): (A6) with and . - œ " \ œ !
If (A6) were replaced with ( ) for , then Lemma 3(b) would become\ / / ¢ \ > - 77> 7ß>"
" " 77ß>" 7> " > for .
Conclusion of proof of Theorem 1. From Lemma 3, . So Lemma 2(c)! µ < œ / /7> 7" 7>"7>"
implies . Therefore,! µ / /"7> 7" 7>
( )! µ \ / /! !Q X7œ" >œ" 7> 7> 7> 7""
œ \ \ /! !Q X7œ" >œ" 7> 7> 7""
( , ) ( , ).œ \ \ / œ \ \! ! !Q X X7œ" >œ" >œ"7> 7> 7 > >" "! !
Thus Lemma 2(a) yields ( , ) and\ µ \!X>œ" > >" !
( , ) ( , ) . _\ ¤ ] Í \ ¤ ] Í \ ¦ ]! ! ! !X X X X>œ" >œ" >œ" >œ"> > > > > > > >" " " "! !
12
Proof of Corollary 2. From the proof of Lemma 3, for each< œ / / µ7> 7" 7>"7>" !
7 œ "ß ÞÞÞßQ > œ #ß $ß µ / / ¸ 1 and ..... From (A5), and the theorem implies ! 7# 7ß>""7>" )
where ( ,..., , / , ,..., ). Then for each and from (A4); so1 œ ! ! ! ! 1 œ 7 >" " " )7# 7ß>" 7>
" "7ß>" 7#>œ ( ) .
Let , . If (A5) is replaced with] œ > − M> )
[( ,..., , , ,...) ( ,..., , , , ,...) ]] ] \ \ µ Ê ] ] \ \ µ" X " # " X " #! !)
then ( ) for all , where / ." " " " " "7> 7X 7 7 7 X" 7X>Xœ > X œ ,
3. Risk Neutrality
Recall that denotes the set of -valued random vectors. A is order-W ? À W Ä‘ ‘Q felicity function
preserving and linear, i.e., it satisfies (9) and (10) for all [ , ], , and :- − ! " E − W F − W
(9) _E ¦ F ?ÐEÑ ?ÐFÑif and only if
(10) )?Ò E Ð" ÑFÓ œ ?ÐEÑ Ð" ?ÐFÑ- - - -
Note that a felicity function is unique up to a positive affine transformation; that is, if and are? ? A
felicity functions, then there are and such that ., ! + − A œ + ,?‘
The following result axiomatizes the comparison of expected present values of cash flows, namely
(6). Let be the subset of whose present value vectors, using any discount factor withZ Z* "
components , are absolutely bounded with probability one.! Ÿ ""7
Theorem 2: (A1), (A2), (A3), (A4), then the following properties areIf satisfies andÐ ¤ ß Z Ñ
equivalent:
(a) (A2 );Ð ¤ ß Z Ñ satisfies -
(b) There exists a felicity function;
(c) , , There exists and such that¯ ‘ " ‘− − > − MQ Q>
(12) X Y ) )¤ Í † IÐ \ † IÐ ]¯ " ¯ "! !X X>œ" >œ"> > > >
for all with for for some , .\ − Z ] − Z \ œ ] œ > X X − M* *> > )
Proof. Since (a) and (b) are immediate consequences of (c) with the felicity function ?ÐEÑ œ
¯ "† IÐ EÑ" , the proof shows that (b) implies (c), and that (a) implies (b). Uniqueness of a felicity
13
function up to a positive affine transformation implies that there is no loss of generality from assuming
?Ð Ñ œ !) .
Lemma 4: (A1), (A2), (A3) ( ) imply:Axioms and and the existence of a felicity function with? ? œ !)
(a) , , , , .?ÐEÑ œ ?Ð EÑ E − W @ÐBÑ œ @Ð BÑ B − @Ð Ñ œ !and with‘ )Q
(b) , , , .?ÐE FÑ œ ?ÐEÑ ?ÐFÑ EßF − W @Ð+ ,Ñ œ @Ð+Ñ @Ð,Ñ +ß , −and ‘Q
Proof: (a): For ,E − W
(Lemma 2(c))E ¸ E Ê E Ð EÑ ¸ Ê Ð"Î#ÑE Ð"Î#ÑÐ EÑ ¸) )
[ ((9))Ê ! œ ? Ð"Î#ÑE Ð"Î#ÑÐ EÑÓ œ Ð"Î#Ò?ÐEÑ ?Ð EÑÓ
.Ê ?ÐEÑ œ ?Ð EÑ
Hence, for all and ( ) .@ÐBÑ œ @Ð BÑ B − @ œ !‘ )Q
(b): From (10),
Ð"Î#ÑE Ð"Î#ÑF ¸ Ð"Î#ÑÐE FÑ Ð"Î#Ñ)
] ]Ê Ð"Î#Ñ?ÐEÑ Ð"Î#Ñ?ÐFÑ œ ?ÒÐ"Î#ÑE Ð"Î#ÑF œ ?ÒÐ"Î#ÑÐE FÑ Ð"Î#Ñ)
œ Ð"Î#Ñ?ÐE FÑ
Let partially order according to first-order stochastic dominance; that is, ifŸ W E Ÿ F
TÖE Á F ! TÖE Ÿ D TÖF Ÿ D× D − B œ ÐB Ñ −} and } for all . Hence, for and‘ ‘Q Q3
C œ ÐC − B Ÿ C B Á C B Ÿ C 3 œ "ß ÞÞÞßQÞ W § W3 3 3Q) , if and for Let be the set of absolutely‘ *
bounded random vectors: : | | , for some .W œ E − W TÖ E O 7 œ "ß ÞÞÞßQ× œ " O _* ˜ ™7 E E
Lemma 5: (A1), (A2), (A3) Axioms and and the existence of an intraperiod utility function with?
? œ !( ) imply:)
(a) , .@ÐBÑ œ † B B −¯ ‘Q
(b) .?ÐEÑ œ † IÐEÑ E − W¯ for all *
Proof: (a): (i) : From Lemma 4(b), .B − M @ÐBÑ œ @Ð B / Ñ œ @ÐB / Ñ œ † BQ Q Q7œ" 7œ"7 7 7 7! ! ¯
(ii) : For each , let / with , . So Lemma 4(b) impliesB − 7 B œ ; = ; = − Mg Q7 7 7 7 7
@ÐB / Ñ œ @Ð; / = Ñ œ ; @Ð/ = @Ð/ Ñ œ @Ð= / = = @ / =7 7 7 7 7 7 7 7 7 7 7 7 7 7 7/ / ). Also, / )= ( / ). Therefore,
@ / = œ @ / = @ B / œ ; @ / = œ B @ÐBÑ œ † B( / ) ( )/ ; so ( ) ( )/ and .7 7 7 7 7 7 7 7 7 7 78 8
14
(iii) : Let with 0 for each . Then Lemma 2(a) andC Ÿ B Ê @ÐCÑ Ÿ @ÐBÑ D œ B C D 77
¯7 7 7 @ÐD / Ñ @ÐB CÑ œ @ÐDÑ @ÐCÑ Ÿ @ÐBÑ0 imply 0. So 0. Therefore, (A2) implies .
(iv) (0, ) : Let , with for each =1,2,... with bothB − _ B B − B Ÿ B Ÿ B 3Q 3 3 Q 3 3+ +g
sequences converging to . From (b) and (c), ) . Letting iB † B œ @ÐB Ñ Ÿ @ÐBÑ@ÐB œ † B Ä _¯ ¯3 3 3 3+ +
implies .@ÐBÑ œ † B¯
(v) : If 0, (d) and Lemma 4(a) imply B − B @ÐB / Ñ œ @Ð B / Ñ‘Q7 7 7 7 7
œ Ð B Ñ@Ð/ Ñ œ B B œ ! @ÐB / Ñ œ @ œ œ B7 7 7 7 7 7 7 7 7¯ ) ¯. If then ( ) 0 .
(b): Part (a) and (11) imply for all . For any , there is a sequence?ÐEÑ œ † IÐEÑ E − W E − W¯ w *
( , ) with , , so ) ) and ) ),E E E E − W E Ÿ E Ÿ E ?ÐE Ÿ ?ÐEÑ Ÿ ?ÐE IÐE Ÿ IÐEÑ Ÿ IÐE3 3 3 3 33 3 w 3 3 3
3 − M IÐE Ä IÐEÑ IÐE Ä IÐE 3 † IÐE Ÿ YÐEÑ Ÿ † IÐE, with ) and ) ). For each , ) ).33 3 3¯ ¯
Convergence of the expectations to implies . IÐEÑ YÐEÑ œ † IÐEѯ
Lemma 6: (A1), (A2), (A2 ), (A3) Axioms and imply the existence of a felicity function.c
Proof. From Herstein and Milnor (1953), (9) and (10) are implied by (13), (14), and (15) (for all
EßFßG − W) which follow:
(13) E ¸ F EÎ# GÎ# ¸ FÎ# GÎ#implies
(14) ;_¦ Wweakly orders
(15) { [ ]: ( } _ _! ! !− !ß " E " ÑF ¦ Ð ¥ ÑG is a closed set
In order to prove (15), let , ) and , ). Then\ œ ÐE F ] œ ÐG F! !
{ [ ]: ( } { [ ]: 0} = { [ ]: } which is_ ~! ! ! ! ! ! !− !ß " E " ÑF ¦ G œ − !ß " \ ] ¤ − !ß " \ ¤ ]
closed due to (A3). Similarly, { [ ]: } is closed._! ! !− !ß " E Ð" ÑF ¥ G
Axiom (A1) implies (14).
In order to prove (13), use (A2), (A2 ), and parts (d) and (e) of Lemma to obtainc
( , ( )( , )_E ¦ F Í EF ¤ Ê "Î# E F ¤!Ñ ! ! !
( )( , ) + ( )( , ) ( )( , Ê "Î# E F "Î# G ¤ "Î# G! ! !Ñ
(( )( ), ) (( )( ), )Ê "Î# E G ¤ "Î# F G! !
( )( ) ( )( )_Ê "Î# E G ¦ "Î# F G
Similarly, ( )( ) ( )( ). _ _ _E ¥ F Í F ¦ E Ê "Î# F G ¦ "Î# E G
15
This completes the proof of Theorem 2.
Extensions of Theorem 2
It is an open question whether there are conditions weaker than (A1) through (A5) and (A2 ) thatc
would axiomatize (4) and the optimization of the “expected utility of the present value" of a time
stream of rewards. See surveys by White (1988) and Whittle (1990).
The remainder of this section concerns several routes that can be followed to extend Lemma 5(b)
beyond (hence Theorem 2 beyond ). Let be the subset of whose elements take only finitelyW Z W W* ‡ w
many values (with probability one). Recall that and denote the zero vector and the - unit) / 77 th
vector in . Suppose that is a felicity function with , for let be a random‘ ) ‘Q QB?Ð † Ñ ?Ð Ñ œ ! B − H
variable with , , , , and . Then (A4)TÖH œ B× œ " @ÐBÑ œ ?ÐH Ñ œ @Ð/ Ñ 7 œ "ß ÞÞÞßQ œ Ð ÑB B 7 7 7¯ ¯ 8
and imply for each . The expected utility formula, a consequence of (10), is@Ð Ñ œ ! ! 7) ¯7
(11) , ?ÐEÑ œ TÖE œ B ×@ÐB Ñ œ IÒ@ÐEÑÓ E − W!3 3 3
w
First, at the expense of additional assumptions and complexity, versions of (11) are valid for larger
collections than [Fishburn (1982)]. These versions would expand the set of random vectors that canWw
be approximated in the proof of Theorem 2. Second, additional assumptions would induce continuity
of in the topology of weak convergence, and that would permit approximating with?Ð † Ñ E − W W*
a sequence in . For a third route, let be the set of random vectors whose components haveW W § W* o
finite expectations. The next result assumes essentially that is continuous at .¦ )
Lemma 7: ( ) Suppose that a felicity function satisfies and? ? œ !)
(16) For all sequences in with as and for allØG Ù W TÖG œ × Ä " 3 Ä _ H − W3 3o o)
, with there is a such that implies andH ¦ 4 3 4 H ¥ G G ¥ H) 3 3
Then and imply for all (A1), (A2), (A3) .?ÐEÑ œ † IÐEÑ E − W¯ o
Proof: (a) From Lemma 5(b), for all there exists with . Let be a sequence# # ! H − W YÐHÑ œ G*3
in with } as . So (16) implies that there exists j such that | | for allW TÖG œ Ä " 3 Ä _ YÐG Ñ o3 3) #
3 4 YÐG Ñ Ä ! 3 Ä _. Therefore, as .3
(b) For , let have component max { ,0}, . ThenE − W E − W 7 E 7 œ "ß ÞÞÞßQo + o th7
E œ E E YÐEÑ œ YÐE YÒ E Ó E Ò E Ó+ + + + + +( ) and ) ( ) with each component of ( )
16
bounded below (above) by 0. So it suffices to prove the assertion for those elements in whoseWo
components are all either bounded above by zero or bounded below by zero.
(c) Let with and 0. Define and if for ;E − W E D H œ E H œ E Ÿ D 7 œ "ß ÞÞÞßQo ) )D 7D
otherwise, and . Hence, for each and } as .H œ H œ E E œ H H D TÖH œ Ä " D Ä _D DD D D) )
So part (a) and Lemmas 4(b) and 5(b) imply ) ) ) )YÐEÑ œ YÐH YÐH œ † IÐH YÐHD DD D8
Ä † IÐEÑ D Ä _ E Ÿ8 ) as . A similar argument suffices for .
Corollary 1 and Lemma 7 extend Theorem 2 to stochastic processes with countable time indices.
COROLLARY 3: Assumptions (8) and (16) imply that Theorem 2 is valid with (12) replaced by (17)
if both expectations exist with finite components:
(17) X Y ¤ Í † IÐ \ Ñ † IÐ ] Ñ \ − Z ß ] − Z8 " 8 "! !_ _>œ" >œ"> > > >
Corollary 2 augments (A1) through (A4) with (A5) to obtain ( ) ; including (A6) yields" "7> 7>"œ
"7 1. So Corollaries 2 and 3 give conditions which imply that preferences are consistent with the
maximization of . The algorithm in Feinberg and Schwartz (1994) computes an¯ "† IÐ \ Ñ!_>œ"
>">
optimal policy for this criterion in a Markov decision process with and finitely many states andQ "
actions. Furukawa (1980) and Henig (1983) study Pareto optimization of the vector IÐ \ Ñ!_>œ"
>">"
when ." "" Qœ † † † œ
4. Scalar Discount Factor
This section combines Corollary 2, Theorem 2, and Koopmans (1972) to specify conditions which
imply . Let be the sequences that are constant with probability one, and for" " ƒ" Qœ † † † œ § Z
B − ² B ² œ =?: 7+B B © B B œ > Xƒ ƒ ƒ ) let | |. Let contain the sequences for which if .> 7 7> X >
Let and with for each and let and withB − C − B ¤ C 8 − B − C −( ) ( ) ( ) ( )8 8 8 8ƒ ƒ \ ƒ ƒ
² B B ² Ä ² C C ² Ä 8 Ä _ ¤( ) ( )8 80 and 0 as . Say that is continuous on with respect toƒ
² † ² B ¤ C if . There are two additional assumptions:
(A3 ) * ¤ ² † ²is continuous on with respect toƒ
(A5 ) ( , , ,...) ( , ,...)c ) \ \ µ \ \ µ" # " #! !implies
Axiom (A5 ) is the converse of (A5).c
17
Proposition 4: (A1) (A5), (A3 ), (A5 )Axioms through and and the existence of an intraperiod utility* c
function imply ." "" Qœ † † † œ
Proof: The assumptions satisfy Postulates 1 through 4 in Koopmans (1972). So in ƒT
(18) B µ C Í ?ÐB Ñ œ ?ÐC Ñ! !X X>œ" >œ"
>" >"> >! !
where is unique and generates a class that is unique up to a positive affine! ‘ ‘− Ð!ß "Ñ ? À ÄQ
transformation. Let and for all ; from Theorem 2, generates a?Ð Ñ œ ! B œ C œ > " ?ÐAÑ œ † A) ) ¯> >
class that is unique up to a positive linear transformation.
Let and ; so from Lemma 3 and (A2). From (18), if\ œ / ] œ / \ µ ] / µ /7> 7> 7" 7> 7> 7"" "
and only if . Thus, . ! ¯ " 8 " !>" >"7 7> 7 7>œ œ
5. Continuity Axioms and Impatience
Reasonable alternative versions of the continuity axiom
(A3) { : } is closed ! !\ ] ¤ Ð £ Ñ !
include these statements:
(A3 ) such that ( , )w \ ¢ Ð ¡ Ñ Ê b ! \ ¡ Ð ¢ Ñ\ a − !! ! ! ! !* *
(A3 ) ww \ ¤ Ð £ Ñ Ê ,\ ¤ Ð £ Ñ a , !! !
Let (A3 ) denote the restriction of (A3) in which .o ] œ !
Proposition 5. (A1), (A2), (A2 ). (A3 ), (A3 ), (A3 ) .Assume and Then and are equivalentc o w ww
Proof. Let (3 ), (3 ), and (3 ) denote the respective unions of {(A1),(A2),(A2 )} with {(A3 )},o c ow ww
{(A3 )}, and {(A3 )}.w ww
(3 ) (3 ): Lemma 2(a) and its proof.o Ê ww
(3 ) (3 ): For , let be a sequence in that converges to from below. To initiate aw ww3Ê , ! Ø, Ù ,g
contrapositive proof, if then (due to (A2 )). So,\ ¡ ,\ ¢! ! c
,\ ¢ , , \ , , 3( ) for sufficiently near (due to (A3 )). Therefore, there exists such that3 3w *
3 3 ¡ ,\ , , \ œ , \ , \ ¡ , \ ¤ 3* implies ( ) . So contrary to for all because! ! !3 3 3 3
\ ¤ , !! and is rational.3
(3 ) (3 ): To initiate a contrapositive proof,ww wÊ
18
(( )( )\ £ \ Í \ \ ¤ Í " \ ¤! ! !! !
Let and ; so . Now (3 ) implies ( )( ) for all . If! ! ! ! !* * " ! " ! , " \ ¤ , !ww !
, œ " ( ) then! "
( )( ) ! ! !£ , " \ Í ¤ \ Í \ £!
(3 ) (3 ): Let with for all . If then ((A1)). If ( ) ww3 3Ê Ä \ ¤ 3 œ ! \ ¤ !o ! ! ! ! ! !! !
then without loss of generality ( ) for all . So (3 ) implies ( ) for all .! !3 "ww ! 3 , \ ¤ £ , !!
Let / to obtain . , œ \ ¤! ! !" !
Impatience
For any and -dimensional random vector , let ( , ) denote the sequence in with first\ − Z Q ^ ^ \ Z
component and component for all . Let be the sequence in with component^ > \ > # \ Z >th th>" #
\ −>" for all t . Versions of impatience that may appear to differ from\
(A6) ( )Sooner is better: \ / / ¢ \- 7> 7ß>"
include the following statements:
(19) ( ) ( ) \ ¢ ] Í \ ] ¢ ^ß\ ^ß ] a ^
(20) ( , ) ( , , ) ( , )\ ¢ ^ \ Í \ ^ \ ¢ ^ \# " #
To interpret (19), which is implicit in Koopmans (1960), suppose that is preferred to . Then \ ] \
is preferred to with an "intensity" which is greater than the intensity with which a delay of is] \
preferred to the same delay of . For an interpretation of (20) that corresponds to the formal definition]
of impatience in Koopmans (1960), suppose that is preferred to the sequence in which is\ \"
replaced by . Then ( , , , ,...) is preferred to ( , , , ,...).^ \ ^ \ \ ^ \ \ \" # $ " # $
Proposition 6. (A1) (A6) (A2 ) (19) (20).Axioms through and imply andc
Proof. To prove the claim regarding (19), let ( ) ( ) have components whose existence is" "œ − !ß "7Q
asserted in Corollary 2, let be the vector with component ( ) , and let be the -vector of" "> >77 Qth 1
ones. Then
( ) ( , ) [ ( )] [ ( )]\ ] ¢ ^ß\ ^ ] Í \ ^ß\ ] ^ß ] ¢ !
( )( ) ( ) ( )Í ¥ \ ] œ \ ]) " " " "! !X X>œ" >œ"
>" > >"> > > >1
19
( ) .Í ¥ \ ] Í \ ¢ ]) "!X>œ"
>"> >
For the claim concerning (20),
( , ) ( )\ ¢ ^ \ Í \ ¦ ^ Í \ ¦ ^# " "1 "
Í \ ^ \ ¦ ^ ^ \" > >X X>œ# >œ#
> >" " " "! ! ( , , ) ( , ) Í \ ^ \ ¢ ^ \" #
6. Attribute Decomposition in Random Vectors
Several of the previous results have interpretations for preference orderings of -dimensionalQ
random vectors. This section is confined to the special case of Theorem 2 when is a binary relation¤
on sequences ( , for all . If axiom (A2 ) were invoked, the restriction would be equivalent toE Ñ E − W! c
fixing and considering the sequences ( ) for all . This yields the following\ − Z Eß\ − Z E − W
version of Theorem 2 in which is regarded as a primitive binary relation on (rather than a binary_¦ W
relation that is induced by on ). Recall that denotes the set of absolutely bounded random¤ Z W*
vectors and consists of vectors whose components have finite expectations.W § Wo
Corollary 4: , ( ), and_If is weakly ordered,Ð ¦ WÑ / ¦ 7 œ "ß ÞÞÞßQ7 )
(21) ( , )_ _EF ¦ Ð Ñ E ¦ Ð ÑF E − W F − W¥ ¥) implies
( , )_Ö − Ò!ß "Ó À E ¦ Ð ÑF× E − W F − W! ! ¥ is closed
then the following properties are equivalent:
(a) , (A2 );_Ð ¦ WÑ satisfies -
(b) There exists a felicity function;
(c) There exists with positive components such that¯ ‘− Q
, _E ¦ F † IÐEÑ † IÐFÑ ÐE − W F − W Ñif and only if ¯ ¯ * *
The assertion above is valid with replaced by if also W W* o
as for all sequences in with and for allØG Ù W TÖG œ × Ä " 3 Ä _ H − W3 3 )
, . with there exists such that implies andH ¦ 4 3 4 H ¥ G G ¥ H) 3 3
20
Assumption (21) corresponds to the decomposition axiom (A2) and is a strong form of utility
independence which is investigated in references cited in §1 and Blackorby, Primont, and Russell
(1978).
7. Conclusion
Assumptions that imply discounting are orthogonal to assumptions that yield an intraperiodnot
utility function. Weak axioms that are consistent with discounting in a deterministic environment carry
the seeds of risk neutrality in a stochastic environment. The following assumptions imply that
preferences among stochastic processes correspond to comparisons of random present value vectors:
weak ordering, decomposition, continuity, and more is better. If there exists an intraperiod utility
function, the same assumptions imply risk neutrality. An axiomatic justification of expected present
value of intraperiod utility (with nonlinear intraperiod utility) seems to require a justification of
discounting in a stochastic environment ( , (7)) with an assumption that is weaker thani.e.
decomposition (A2).
Acknowledgements
The author is grateful to Professor James C. Alexander for comments and the argument on which
the proof of Lemma 3(a) is based, to Professor Eugene Feinberg and Dr. Charu Sinha for comments,
and to an anonymous commentator on an earlier version of the paper who noted that confining to ] !
in (A3) permitted a counterexample to a lemma.
21
References
Altug, S. and Labadie, P. , Academic Press, New York, 1994. Dynamic Choice and Asset Markets
Arrow, Kenneth J., Karlin, Samuel and Scarf, Herbert. Studies in the Mathematical Theory of
Inventory and Production, Stanford University Press, Stanford, CA, 1962.
Balcer, Yves and Lippman, Steven A. “Technology Expectations and Adoption of Improved
Technology." , 1984, 34, pp. 292-318.J. Economic Theory
Blackorby, C., Primont, D., and Russell, R. R. Duality, Separability, and Functional Structure:
Theory and Economic Applications. North-Holland, New York, 1978.
______________________________________ "Separability: A Survey." Handbook of Utility Theory,
Volume 1: Principles. Kluwer Academic Publishers, Dordrecht, 1998.
Blair, Charles E. “Axioms and Examples Related to Ordinal Dynamic Programming." Mathematics
of Operations Research, 1984, 8, pp. 345-347.
Cochrane, John H. (2000). , Princeton University Press, Princeton, NJ.Asset Pricing
Denardo, Eric V. “Contraction Mappings in the Theory Underlying Dynamic Programming." SIAM
Review, 1967, 9, pp. 165-77.
Duffie, Darrell. , 3 ed., Princeton, NJ: Princeton University Press,Dynamic Asset Pricing Theory rd
2001.
Dyer, James S. and Sarin, Rakesh “Measureable Multiattribute Value Functions." Operations
Research, 1979, 27, pp. 810-22.
Epstein, Larry G. “Stationary Cardinal Utility and Optimal Growth under Uncertainty." J. Economic
Theory, 1983, 31, pp. 133-151.
Farquhar, Peter H. “A Survey of Multiattribute Utility Theory and Applications." pp. 59-89 in
Multiple Criteria Decision Making, Starr, M. K. and Zeleny, M. eds., Amsterdam: North-Holland,
1977.
Feinberg, Eugene and Schwartz, Adam. “Markov Decision Models with Weighted Discounted
Criteria," , 1994, 19, pp. 152-168.Mathematics of Operations Research
22
Fishburn, Peter C. . Huntington, NY: Robert E. KriegerUtility Theory for Decision Making
Publishing Co., 1970.
_______________. “A Survey of Multiattribute/Multicriterion Evaluation Theories," pp. 181-224 in
Multiple Criteria Problem Solving, Zionts, S. ed., Springer-Verlag, Berlin, 1978.
_______________. . Dordrecht, Holland: D. Reidel, 1982.The Foundations of Expected Utility
_______________ and Rubinstein, Ariel. "Time Preference." 1982,International Economic Review,
23, 2, pp. 677-94.
Frederick, Shane, Loewenstein, George, and O'Donoghue, Ted. "Time Discounting and Time
Preference: A Critical Review." , 2003, to appear.J. of Economic Literature
________________ , Dordrecht, Holland: D. Reidel, 1982.The Foundations of Expected Utility
Furukawa, N. “Characterization of Optimal Policies in Vector-Valued Markovian Decision
Processes," , 1980, 5, pp. 271-79.Mathematics of Operations Research
Henig, Mordecai I. “Vector-Valued Dynamic Programming." , 1983,SIAM J. Control & Optimization
21, pp. 490-99.
Herstein, Israel N. and Milnor, John W. “An Axiomatic Approach to Measureable Utility."
Econometrica, 1953, 21, pp. 291-297.
Ho, T. S. Y. and S-B. Lee, S-B. “Term Structure Movements and Pricing Interest Rate Contingent
Claims," , 1986, XLI, pp. 1011-29.J. Finance
Johnson, T. H. and Donaldson, John B. “The Structure of Intertemporal Preferences under
Uncertainty and Time Consistent Plans," , 1985, 53, pp. 1451-58.Econometrica
Jorgenson, Dale W., McCall, John J. and Radner, Roy. , Amsterdam:Optimal Replacement Policy
North-Holland, 1967.
Keeney, Ralph L. and Raiffa, Howard. , New York: John Wiley,Decisions with Multiple Objectives
1976.
Koopmans, Tjallings C. “Stationary Ordinal Utility and Impatience," , 1960, 28, pp.Econometrica
287-309.
23
Koopmans, Tjallings C., Diamond, Peter. A., and Williamson, R. E. “Stationary Utility and Time
Perspective," , 1964, 32, pp. 82-100.Econometrica
Koopmans, Tjallings C. “Representation of Preference Orderings Over Time." Chap. 4 in Decisions
and Organizations, McGuire, C. B. and Radner, R. eds., , Amsterdam: North-Holland, 1972.
Kreps, David M. and Porteus, Evan L. “Temporal von Neumann-Morgenstern and Induced
Preferences," , 1979, 20, pp. 81-109.J. Economic Theory
Lancaster, Kelvin J. “An Axiomatic Theory of Consumer Time Preference." International Economic
Review, 1963, 4, pp. 221-31.
Machina, Mark J. “Dynamic Consistency and Non-Expected Utility Models of Choice." J. Economic
Literature, 1989, 27, pp. 1622-68.
McNamara, John M. "The Policy which Maximises Long-Term Survival of an Animal Faced with
the Risks of Starvation and Predation." , 1990, 22, pp. 295-308.Advances in Applied Probability
Mendelssohn, Roy. “Using Markov Decision Models for Purposes other than Simple Optimization:
Analyzing the Consequences of Policy Alternatives on the Management of Salmon Runs."
Fisheries Bulletin, 1979, 78, pp. 35-50.
Meyer, Richard F. "Preferences Over Time." Chapter 9 in Keeney, Ralph L. and Raiffa, Howard,
Decisions with Multiple Objectives: Preferences and Value Tradeoffs, New York: Wiley, 1976.
Miyamoto, J. M. and Wakker, P. "Multiattribute Utility Theory without Expected Utility
Foundations." , 1996, 44, pp. 313-326.Operations Research
Monahan, George E. P. “Optimal Advertising with Stochastic Demand." , 1983,Management Science
29, pp. 106-17.
Moriarity, S. and Allen, C. P. , New York: Harper and Row, 1984.Cost Accounting
Murphy, F. H., Toman, M. A.and Weiss, H. J. “A Dynamic Nash Game Model of Oil Market
Disruptions and Strategic Stockpiling." , 1989, 37, pp. 958-71.Operations Research
Rosenthal, Robert. Personal communication, Stony Brook, NY, 1987.
Samuelson, Paul A. "A Note on Measurement of Utility." , 1937, 4, pp.Review of Economic Studies
155-161.
24
Sethi, Suresh P. "Optimal Consumption-Investment Decisions allowing for Bankruptcy: A Survey,"
pp. 397-426 in , Ziemba, W. T. and Mulvey, J.M. eds.,Worldwide Asset and Liability Modeling
Cambridge, UK: Cambridge University Press, 1998.
Shapley, Lloyd S. "Stochastic Games." , 1953, 39, pp.Proc. National Academy of Sciences, USA
1095-1100.
Sharpe, William F. , 3rd ed.,Englewood Cliffs, NJ: Prentice-Hall, 1985.Investments
Shubik, Martin. "Game theory, behavior and the paradox of the prisoner's dilemma: Three solutions."
J. Conflict Resolution, 1970, 14, pp. 181-93.
Shubik, Martin and Sobel, Matthew J. "On Matching Book: A Problem in Banking and Corporate
Finance." , 1992, 38, pp. 827-39.Management Science
Shubik, Martin and Thompson, Gerald L. "Games of Economic Survival." Naval Research
Logistics Quart , 1959,. 6, pp. 111-23.
Sobel, Matthew J. “Ordinal Dynamic Programming," , 1975, 21, pp. 967-75.Management Science
Sobel, Matthew J. “Ordinal Sequential Games." , 1980, XIV, pp. 1571-82.Économies et Sociétés
von Neumann, J. and Morgenstern, O. , Princeton, NJ:Theory of Games and Economic Behavior
Princeton University Press, 1944.
Wakker, P. P. "Additive Representations on Rank-Ordered Sets II. The Topological Approach." J.
Mathematical Economics, 1993, 22, pp. 1-26.
White, Douglas J. “Mean, Variance, and Probabilistic Criteria in Finite Markov Decision Processes: A
Review." , 1988, 56, pp. 1-29.J. Optimization Theory & Applications
Whittle, Peter C. , Wiley, New York, 1990.Risk-Sensitive Optimal Control
Williams, Albert C. and Nassar, J. I. “Financial Measurement of Capital Investments." Management
Science, 1966, 12, pp. 851-64.
Yeh, William W-G. “Reservoir Management and Operations Models: A State-of-the-Art Review,"
Water Resources Research, 1985, 21, pp. 1797-1818.