NOT FOR QUOTATION WITHOUT THE PERMISSION OF THE AUTHOR
A Prescriptive Model for Averse-Prone Risk Attitudes
C h a r l e s
M.
H a r v e yOctober
1986 WP-86-70Working P a p e r s are interim r e p o r t s on work of t h e International Institute for Applied Systems Analysis and have received only limited review. Views o r opinions expressed herein do not necessarily r e p r e s e n t those of t h e Institute or of its National Member Organizations.
INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS A-2361 Laxenburg, Austria
Foreword
This paper presents
a
prescriptive model f o ra
decision maker's risk attitude toward financial outcomes that have important non-monetary effects, f o r example, effects on how the decision maker is judged by himself and by others. The model represents the risk attitude ofa
decision maker who is risk averse in the absence of such psychological effects, but who is risk prone in their presence f o r actions leadingto
net losses o r the status quo. The model is examined f o r its adherenceto
normative principles. In particular, it is argued that the principle of dominance should be specified without any assumptions on preferences between conjunctions of lotteries; such assumptions a r e shownto
imply the apparently stronger princi- ple of risk neutrality.Alexander B. Kurzhanski Chairman
System and Decision Sciences Program
A Prescriptive Model for Averse-Prone Risk Attitudes Charles
M.Harvey
Introduction
Behavioral studies of people's r i s k attitudes h a v e found t h a t in a v a r i e t y of c o n t e x t s t h e majority of people are r i s k a v e r s e in t h e i r p r e f e r e n c e s among
ac-
tions leadingto
gainsor
t h e s t a t u s quo but are r i s k p r o n e in t h e i r p r e f e r e n c e s among actions leadingto
lossesor
t h e s t a t u s quo. Such p r e f e r e n c e s will b ere-
f e r r e dto
in t h i s p a p e ras
anaverse-prone r i s k a t t i t u d e .
During t h e e a r l y development of e x p e c t e d utility t h e o r y , Friedman and Savage (1948) and Markowitz (1952) discussed averse-prone r i s k a t t i t u d e s (and even t h r e e - and four-piece r i s k attitudes). Relevant empirical work includes five stu- dies, Barnes a n d Reinmuth (1976), Grayson (1960). Green (1963). H a l t e r a n d Dean (1971), and Swalm (1966), t h a t
were
examined by Fishburn and Kochenberger (1979), a n d were reexamined t o g e t h e r with o t h e r empirical studies by Hershey, Kunreuther, and Schoemaker (1982). Other empirical work and analysesare
in Dickson (1981), Fuchs (1976), Hershey and Schoemaker (198Oa), Langhhunn, Payne, and Crum (1980), Schoemaker and Kunreuther (1979), Wehrung,
Bassler, MacCrimmon, and S t a n b u r g (1978), Wehrung,
MacCrimmon, and B r o t h e r s (1984), and Williams (1966).This p a p e r is
a
modeling a n d a subsequent analysis of averse-prone r i s k atti- tudes. In t h e c u r r e n t milieu of r e s e a r c h on p r e f e r e n c e models, i t i s importantto
emphasize t h a t t h e models p r e s e n t e d h e r eare
p r e s c r i p t i v e r a t h e r t h a n normativeor
descriptive. Theyare
intendedto
help a decision maker whose a c t u a l p r e f e r - ences, i.e., t h e p r e f e r e n c e s lying behind his cognitive limitations,are
such t h a t h e h a s a n averse-prone r i s k attitude. These models may o r may not conform to vari- o u s assumptions of rationality (a m a t t e r t h a t is examined in t h e second p a r t of t h i s p a p e r ) . Moreover, t h e y mayor
may not provide a n a c c u r a t eor
predictive model- ing of t h e behavior of most people ( a m a t t e r t h a t i s commented on b u t it is nottest-
e d in t h i s p a p e r ) . Bell, Raiffa, a n d Tversky (1984) and Schoemaker (1982) contain r e c e n t , g e n e r a l discussions of t h e distinctions between t h e normative, descriptive, and p r e s c r i p t i v e modeling of p r e f e r e n c e s .Section 1 discusses t h e types of psychological responses
to
financial gains and losses t h a t may inducean
averse-prone risk attitude, and p r e s e n t s a r a t h e r general model of such r i s k attitudes. Section 2 then discusses a specialization of t h i s model in which t h e psychological effects of gains and lossesare
priced-outas a
"reward" t h a t i s independent of t h e magnitude of t h e gain and a "penalty" t h a t is independent of t h e magnitude of t h e loss. In t h i s model, anyone who i s r i s k a v e r s e when t h e psychological e f f e c t sare
absent will havean
averse-prone r i s k attitude o v e r s o m e m g e of monetary changes when t h e psychological e f f e c t sare
present.Sections 3 and 4 examine which normative principles
are or
are not violated by a n averse-prone r i s k attitude. The salient conclusion is not a list of yes/noanswers
but t h a t c e r t a i n conditions implicit in t h e common normative principles needto
b e made explicit and then examined in isolation. In particular, i t is shown t h a t c e r t a i n conditions on p r e f e r e n c e s between conjunctions of lotteries t h a tare
often implicit in t h e dominance principle imply t h e apparently s t r o n g e r principle of r i s k neutrality. I t is argued t h a t these conditions on conjunctions (and a condi- tion implicit in t h e framing principle) should not b e included as normative princi- ples.Section 5 then r e t u r n s
to
a prescriptive viewpoint and'discusses t h e potential uses and abuses of a n averse-prone r i s k model f o r a decision analysis application in which r i s k aversion f o r gains but r i s k proneness f o r losses isan
important p r e f e r e n c e issue.1.
P s y c h o l o g i c a l - E f f e d sY
odelaS e v e r a l r e s e a r c h e r s have suggested t h a t
a
decision maker's r i s k attitude may depend on various psychological effects of financial outcomes t h a t might b e includ- e d as p a r t of the description of t h e decision maker's possible consequences. For example, P e t e r Fishburn, Ralph Keeney, and Richard Meyer (in t h e discussion fol- lowing a p a p e r presented by Tversky, 1977) suggest t h e use ofan
additionalattri-
bute to explain r i s k p r o n e p r e f e r e n c e s f o r potential losses in financial choices. A similar idea is suggested in Keeney (1984) by his argument t h a t t h e ethics ofa
deontological moralist need not violate t h e expected utility conditions provided t h a t t h e social consequences of policy decisionsare
adequately defined (p. 122).Moreover, Raiffa (1984) discusses psychological effects by imagining t h a t t h e deci- sion maker has a n e x t e r n a l or internal kibitzer whose r e m a r k s could b e used
to
provide a
m o r e
sophisticated description of t h e consequencesto
t h e decision maker.In t h i s section, a p r e f e r e n c e model is developed t h a t is consonant with t h e above ideas. Suppose t h a t financial outcomes t o t h e decision maker
are
measured by a variablez
such t h a tz >
0 r e p r e s e n t s n e t gains,z <
0 r e p r e s e n t s n e t losses, andz =
0 r e p r e s e n t s t h e s t a t u s quo. An individual who i s makinga
decision t h a t a p p e a r sto
depend only on t h e financial amountsz
may be influenced by t h e antici- pation of his psychological responsesto
t h e possible consequences ofa
chosen ac- tion. To d e s c r i b e t h e s e effects, it will b e helpful t o distinguish betweena
decision maker who is acting on his own behalf anda
decision maker who acting as a n agent f o r a n organization.A person who is acting on his own behalf, e.g., a private e n t r e p r e n e u r o r a n individual investor, may feel embarrassment
or a
loss of self-esteem ifa
financial loss should occur. This person may also feel p r i d eor
increased self-esteem asa
r e s u l t ofa
financial gain. Such a t y p e of psychological response i s h e r e dis- tinguished from a change in financial position. t h a t is, from t h e event t h a t t h e per- son will havea
c e r t a i n l e s s e r or g r e a t e r bmountto
spend o v e r his lifetime.For a person who i s acting
as
a n agent f o ran
organization. e.g., a business ex- ecutiveor
a government administrator. t h e r eare
e x t e r n a l p r e s s u r e s in additionto
t h e type of ego involvement described above. The person wishes
to
b e favorably judged by thoseto
whom h e i s accountable; f o r example, t h e person's primary con- c e r n may b eto
maintainor to
enhance his reputation.A s
a
primary assumption in t h i s p a p e r , suppose t h a t t h e psychological effectsto
b e considered can b e measured as components of t h e consequencesto
t h e deci- sion maker. For modeling purposes, t h e s e types of psychological effects will b e calledm c t s
attributes. Each effects a t t r i b u t e will b e measured by a variable zf.
i=
1 ,..., n .
Consequences will b e described by both t h e variablez
f o r t h e monetary a t t r i b u t e and t h e variables zl,..., z,
f o r t h e effects attributes; thus.each consequence will b e denoted by
a
v e c t o r ( z,z I,...
,z, ).By contrast, Bell (1982). (1983). (1985) has modeled psychological responses such
as
"regret" and "disappointment" t h a t depend upon t h e decision maker's per- ception of a l l of t h e available l o t t e r i e sor
of t h e e n t i r e lottery t h a t i s chosen. A s s t a t e d in Bell (1982) "regret i s measured. . . as
t h e difference in value between t h eassets
actually received and t h e highest level ofassets
produced by other al- ternatives [italics Bell's]."
In
a
specific decision context, t h e amounts z l ,..., z,
may b e highly c o r r e l a t e dto
t h e financial amountz
and may even b e functionally dependent onz .
However, i t will b e assumed t h a t t h e decision maker i s familiar witha
sufficiently wide v a r i e t y of c o n t e x t s such t h a t his p r e f e r e n c e scan
b e considered ona
p r o d u c tset
of potential consequences ( z ,z l,... ,z, ) where t h e variablesz
, z l,... ,z,are
defined on specific intervals.(A) Suppose t h a t t h e decision maker's t r a d e o f f s satisfy t h e willingness-to-pay conditions (see, e.g., Keeney and Raiffa, 1976, pp. 125-127, and Harvey, 1985).
Then, p r e f e r e n c e s among consequences in t h e product
set can
b e r e p r e s e n t e d bya
value function of t h e formHere, t h e amounts gi (zi ),i
=
1,...,
n , c a n b e assessed as p r i c i n g - o u t a m o u n t s f o r t h e psychological responses; t h a t is, f o r some specified r e s p o n s ezi *
of t h e i-th e f f e c t s a t t r i b u t e , gi (zi) i s t h a t amount such t h a t t h e decision maker would just b e willing t o pay gi (zi) in o r d e rto
obtainzi
r a t h e r t h a nzi * .
The value V of a conse- quence ( z , z l ,...,
z,) c a n b e i n t e r p r e t e d as t h e financial amount such t h a t t h e consequence ( z ,z
l,... , z,
) i s indifferentto
t h e consequence (V,z l*,...
,z,*
).(B) Suppose t h a t t h e decision maker's r i s k a t t i t u d e satisfies t h e conditions of expected utility. Then, p r e f e r e n c e s among l o t t e r i e s c a n b e r e p r e s e n t e d by
a
utili- t y function of t h e formH e r e , t h e function
u
c a n b e i n t e r p r e t e d as a conditional utility function on finan- cial amountsz
given t h a t t h e psychological e f f e c t sare
t h e specified amounts zl*,...,
Zn*.
Consider n e x t t h e causal relations between receiving a n e t gain or a n e t loss
z
i s a specific c o n t e x t and t h e resulting psychological effects. Suppose t h a t in any o n e decision context t h e psychological e f f e c t s described byzl. ..., z, are
func- tionally dependent on t h e financial outcomez .
Then, conditional on a given con- text:zi = ri
( z ) , i=
1 ,...,
n , f o r some functionsr*.
The functionsri
w i l l b e re- f e r r e dto as
r e s p o n s e j b n c t i o n s . Note t h a ta
decision context does not a f f e c t t r a d e o f f s between t h e a t t r i b u t e s b u t r a t h e r r e s t r i c t s t h e domain of potential consequences.For prescriptive purposes, i t will be useful
to
compare t h e following two types of decision contexts.(i)
The a c t s contezt:
Here t h e decision maker anticipates t h a t his psycho- logical responses are importantto
him and should b e included in describing t h e consequences of his actions.(C) Suppose t h a t t h e response functions zi
= ri
( z ), i= 1 , ... ,n , as
discussed above denote t h e decision maker's psychological responses in t h e effects context.Suppose t h a t l a r g e r monetary amounts z lead
to
responses zi= ri
( z ), i= 1 , ... ,n ,
that areat
leastas
preferred. The responsesto
maintaining t h e status quo, i.e.,to
z
= 0 ,
will be denoted by zi* = ri (0)'
i= 1 ,..., n .
The amounts zi*,
i= 1 ,..., n ,
will b e calledstandard a e c t s .
(ii) The
no-effects contezt:
Here t h e psychological responses in t h e effects context are e i t h e r absentor
are omitted f r o m consideration.( D ) Suppose t h a t in t h e no-effects context the psychological responses
to
any financial change zare
t h e standard effects zi*
, i=
1,..., n .
Thus, t h e effects of any change z in t h e no-effects contextare
t h esame as
t h e effects of no change, z= 0 ,
in t h e effects context.Definition
1.
A preference modelas
described in conditions ( A ) - ( D ) will b e calleda p ~ a j c h o l o g i c a Z ~ e c t s model.
Any utility functionu ( z )
defined on monetary amounts z in t h e no-effects context will be called aho-q#kcts u t i l i t y function;
any utility function
w
( z ) defined on monetary amounts z in t h e effects context will be called a nsects u t i l i t y &nction.
For
a
net gainor
loss z , consider t h e associateda c t s amount e
( z ) defined bywhere
ri
( z ), i=
1,... ,n ,
are t h e psychological effects of z in t h e effects context.The monetary amount e ( z ) can be interpreted
as
t h a t amount such t h a t t h e conse- quence( 0 , z l,...
,zn ) i s indifferentto
t h e consequence( e
( z ) , z....
,zn*
), t h a t is,e ( z )
i s t h e total pricing-out amount of t h e psychological effects zi= r i ( z ) ,
i= 1 ,..., n .
Theorem
1.
For a psychological-effects model, ifu
( z ) i s any utility function for t h e decision maker's risk attitude in t h e no-effects context and w ( z ) i s any utility function f o r t h e decision maker's risk attitude in t h e effects context, thenf o r
s o m e
normalization constants a>
0 and 6 .The psychological-effects model discussed above is f a r
too
r e s t r i c t i v eto
describe t h e heuristic biases t h a t might b e responsible forat
least p a r t of a n ob- served averse-prone r i s k attitude. The model is intended as a possible formulation of t h a t p a r t of a n averse-prone r i s k attitude which t h e decision maker r e g a r d sas
dueto
his underlying preferenoes. For sucha
prescriptive purpose, i t i s usefulto
f u r t h e r specialize t h e model; t h e following section presents one means of doing so.2. The Rermrd-Penalty Podel
This section discusses
a
special type of psychological-effects model f o rcom-
paring a decision maker's p r e f e r e n c e s in t h e effects context and in t h e no-effects context. The model is intendedto
b e sufficiently specificto
b e t r a c t a b l e for deci- sion analysis applications.As a strong causal assumption, suppose t h a t in the effects context t h e psycho- logical responses are constant in t h e sense t h a t
f o r some constant amounts zi+ and zi-, i
=
1,..., n.
Thus, t h e effectszi
in t h e ef- f e c t s context depend only on whether t h e financial change is a gain or a loss, and not on t h e magnitude of t h e gainor
loss. This type of dependence may by a p p m p r i -ate as
a modeling simplification f o r a variety of decision situations. For example, a decision maker actingas an
agent may believe t h a t h e is being judged in p a r t ina
superficial manner by whether h e succeeds ( z>
0), maintains t h e status quo ( z=
0),or
fails ( z<
0).I t follows directly from equations (3) and
(5)
t h a t r i ( z ) = 'I
p f o r z > O e ( z ) = O f o r z = O-n f o r
z <
0zi+
f o r z > O2; f o r z = O
zi-
f o rz <
0b
f o r some constant financial amounts p 2 0 and n 2 0. The tradeoffs amounts p and
n
can b e r e g a r d e das
t h e e x t r a "reward" and "penalty" in t h e effects context ofa
gain and a loss respectively.Definition 2.
A psychological effects model t h a t satisfies t h e condition of con- s t a n t psychological e f f e c t s summarized in (5), (6) will b e calleda
reward-penalty model.Figure 1 illustrates
a
reward-penalty model in which t h e utility function w (z) f o r t h e effects context correspondsto
a utility function u (z) for t h e no-effects context t h a t r e p r e s e n t s r i s k aversion. In t h e effects context, t h e penaltyn >
0leads
to
r i s k proneness among l o t t e r i e s whose possible consequencesare
n e t losses and t h e s t a t u s quo; t h e reward p>
0 leadsto
g r e a t e r risk aversion among l o t t e r i e s whose possible consequences are net gains and t h e s t a t u s quo. These p r o p e r t i e s a r e illustrated by t h e dotted lines in Figure 1. Thus, w ( z ) r e p r e s e n t s a n averse-prone r i s k attitude o v e r a r a n g e of amounts z including both n e t gains and n e t losses.Figure 1. A Utlllty Functlon for the EfYeats Context
A numerical illustration of
a
reward-penalty model can b e calculated by con- sidering t h e following well-known choice problem presented in Tversky and Kahne- man (1981). The p e r c e n t a g e s noted by t h e a l t e r n a t i v e actions are based on t h e responses of 150 studentsat
Stanford University and t h e University of British Columbia.Problem 1. Imagine t h a t you f a c e t h e following p a i r of c o n c u r r e n t decisions.
F i r s t examine both decisions, then indicate t h e options you p r e f e r .
Decision (i). Choose between:
A. a sure gain of $240 [84 p e r c e n t ]
B. 25% chance
to
gain 8 l 0 0 0 , and 75% chanceto
gain nothing116
p e r c e n t ]Decision (ii). Choose between:
C.
a
s u r eloss
of S750 [13 p e r c e n t ]D.
75% c h a n c eto
lose 8 l 0 0 0 , and 25% chanceto
lose nothing [87 percent].A majority [73 p e r c e n t ] of t h e respondents chose actions A and D. The ex- pected monetary values of t h e f o u r actions are: E ( A )
=
8240<
E @ )=
8250, andE ( C )
=
-S750=
E(D). Thus, t h e students were r i s k a v e r s e in decision (i) but r i s k p r o n e in decision (ii).To model t h e s e p r e f e r e n c e s with
a
reward-penalty model, suppose t h a t in t h e no-effects context a p e r s o n h a s constant r i s k aversion ( a t least o v e r t h e r a n g e of gains and losses considered) andassesses a
certainty equivalent of $245 f o r t h e l o t t e r y B. Then, in t h e no-effects context, t h e person i s mildly r i s k a v e r s e ; h e p r e f e r s Bto
A and p r e f e r s Cto
D ( t h e opposite of t h e p r e f e r e n c e s observed by Tversky and Kahneman).Now, consider t h e person's p r e f e r e n c e s in t h e e f f e c t s context. F o r
a
reward amount of p= S7 or
more, t h e p e r s o n will become sufficiently r i s k a v e r s e among gains l o t t e r i e s so t h a t h e p r e f e r s Ato
B. Fora
penalty amount of rr=
$21 or more, t h e person will become sufficiently r i s k p r o n e among losses l o t t e r i e s so t h a t h e p r e f e r s Dto
C. Therefore, psychological responsesto
gains andlosses
t h a t leadto
only modest t r a d e o f f s amounts p andn are
sufficientto
induce t h e choice behavior observed by Tversky and Kahneman.The assumptions of a reward-penalty model imply in general a number of pro- perties f o r t h e decision maker's preferences in t h e effects context. These pro- perties are listed in t h e result below. Here, a lottery having a net gain
or
loss ofz
with probability p anda
net gainor
loss ofz'
with probability q=
1 - p i s denoted by<
p $2, q Sz '>.
Theorem
2. Suppose t h a ta
decision maker's preferences satisfy t h e conditions of a reward-penalty model.Part
L If in t h e no-effects context t h e decision maker is risk a v e r s e , then in the effects context his preferences will have t h e following properties:(a) If t h e penalty K is positive, then preferences
are
risk prone f o r moderate losses, t h a t is, f o r any probability 0<
p<
1 ,<
p Sz,
q 80>
i sp r e f e r r e d
to
t h e amount Spz f o r sufficiently small negativez.
(b) Preferences
are
risk a v e r s e f o r moderateor
l a r g e gains, t h a t is, for any probability 0<
p<
1 , t h e amount Spz i s p r e f e r r e dto <
p $2, q 80>
f o rany positive
z .
(c) If K i s g r e a t e r than p, then preferences
are
r i s k a v e r s e f o r symmetric lotteries, t h a t is, 80 is p r e f e r r e dto <
$2,+
S ( z )>
f o r any non-zero amountz .
Part II.
If in t h e no-effects context t h e decision maker has decreasing risk aver- sion and has unbounded utility from below (i.e., f o r any amountsz < z'
t h e r e is an amountz "< z
such t h a tz
i s p r e f e r r e dto <
Sz ",+
Sz'>
), then in t h e effects context his preferences have t h e following additional properties:(d) P r e f e r e n c e s
are
risk a v e r s e f o r l a r g e losses, t h a t is, f o r any probability 0<
p<
1, Spz is p r e f e r r e dto <
p $2, q SO>
f o r sufficiently l a r g e negativez .
(e) If K
>
0, then preferencesare
more r i s k a v e r s e in t h e effects context then in t h e no-effects context for s t r i c t losses, t h a t is, f o r anyt w o
s t r i c t lossesz
, z '<
0, t h e certainty equivalent of<
p $2, q $2'>
i s less in the effects context than in t h e no-effects context.(f) If p
>
0, then preferencesare less
risk a v e r s e in t h e effects context than in t h e no-effects context f o r s t r i c t gains, t h a t is, for anyt w o strict
gainsz
, z '>
0, t h e certainty equivalent of<
p $2, q $2'>
ismore
in t h e effects context than in t h e no-effects context.The purpose of Theorem 2 i s not
to
provide a means of testing t h e descriptive a c c u r a c y of a reward-penalty model. The intent i s r a t h e rto
make explicitto a
de- cision maker some of t h e implications of adopting such a p r e s c r i p t i v e model.3. Conformity w i t h Normative Principlw
This section examines whether
a
r i s k attitude in a reward-penalty model, or a averse-prone r i s k a t t i t u d e in general, confirmsto
c e r t a i n conditions on p r e f e r - ences t h a t have been r e g a r d e d from a normative viewpoint as principals f o r ra- tional decision making. The violation of t h e s e principles by people's choice behavior in non-transparent decision problems h a s been w e l l documented, f o r ex- ample, in t h e work of D. Kahneman and A. Tversky (1979), (1981), and (1986).3.1.
T r a n s i t i v i t y , Dominance, a n d Independence..
F o r any reward-penalty model in which t h e r e i s r i s k aversion in t h e no-effects context anda
non-zero penaltyn,
t h e r e will b e in t h e e f f e c t s context a n averse-prone r i s k a t t i t u d e as described in p a r t s (a), (b) of Theorem 2. Such p r e f e r e n c e s
are
r e p r e s e n t e d bya
utility func- tion w ( z )as
in (4), and henceare
consistent with t h e normative principles of ex- pected utility. I t follows t h a t , in p a r t i c u l a r , t h e s e averse-prone r i s k attitudes satisfy t h e principles of transitivity, stochastic dominance, and independence.3.2. f i a m i n g Consistency.
Consider any averse-prone r i s k attitude with r e s p e c tto
a variablez.
In t h i s subsection, t h e r i s k attitude mayor
may not confirmto
t h eprinciples of e x p e c t e d utility. Let I
= <
pi ,zi>
denote alottery
having possible consequenceszi
with probabilities pi where i=
1 ,...
,m.
The v a r i a b l e
z
is intendedto
measure changes in monetary position. i.e., ei- t h e r changes in t h e decision maker's personal financesor
changes in t h e finances ofan
organization f o r which t h e decision maker is acting as a n agent. The conse- quencesto
t h e decision maker canalso
b e describedas
finalasset
positions meas- u r e d by a v a r i a b l e y . Assume that: (1)a
c u r r e n tasset
positionc
c a n b e defined (but not necessarily evaluated), (2) t h e amountsz are
n e t monetary gains and losses, e.g., n e t p r e s e n t valuesor
n e t after-tax profits, and (3) t h e finalasset
po- sition y implied byc
andz
is specified by t h e formula y=
c+ z.
For a situation in which
z
measuresrates
of r e t u r n o r net pre-tax profits. t h e formula relating y t oz
f o r a fixed c u r r e n t asset position c will differ from y=
c+ z
, but t h e r e will still b e a one-to-one relationship y= f (z I
c ) between financial changesz
and finalasset
positions y . The r e s u l t s in t h i s subsection easily generalizeto
such situations.The p r e f e r e n c e relation concerning financial changes
z
conditional on a c u r r e n t asset position c will b e denoted by & ~ c . An associated preference rela- tion&
concerning finalasset
positions y is defined bywhere y i
=
c+ zt .
y j= c + zj
f o r i=
1,..., m
and j=
1,..., m '.
The equivalence (7) between p r e f e r e n c e s when outcomesare
described by net gainsor losses
and p r e f e r e n c e s when outcomes are described by asset positions will be called afram-
i n gtranqfomation.
Definition 3. A decision maker's p r e f e r e n c e s such t h a t t h e framing transforma- tion (7) is satisfied f o r every c u r r e n t
asset
positionc
will b e calledframing con- sistent.
The condition of framing consistency can also b e regarded
as
a consistency .condition on p r e f e r e n c e s concerning financial changes f o r different c u r r e n t asset positions c and c ' . More precisely, framing consistency implies t h e condition t h a t
where c , c f a r e any two c u r r e n t
asset
positions and d=
c-
c ' . Conversely, t h e condition (8) implies t h a t if a p r e f e r e n c e relation&,
on final asset positions i s de- fined by (7) with a specific amount c , then (7) is also satisfied f o r any o t h e r m o u n tc'.
Kahneman and Tversky (1979, p. 273) have demonstrated t h a t t h e manifest behavior of students who are presented with hypothetical choice problems is not in accord with t h e condition of framing consistency; often t h e r e a r e r e v e r s a l s in preference depending on whether t h e problem is framed in
terms
ofa
c u r r e n t as- s e t position c o r in terms of a n o t h e r c u r r e n tasset
position c ' .Theorem 3. The condition t h a t t h e preference relations & ~ c are averse-prone risk attitudes f o r more than one c u r r e n t
asset
position c i s inconsistent with t h e principle of framing consistency.The r e s u l t identifies a conflict between
a
p r e f e r e n c e condition havinga
strong normative appeal and t h e choice behavior r e p o r t e d in many a r t i c l e s , e.g., Fuchs (1976), Green (1963). G r e t h e r and Plott (1979). Halter and Dean (1971), Hershey and Schoemaker (1980a), Hershey and Schoemaker (1980b), Kahneman and Tversky (1979), Langhhunn, Payne, and Crum (1980), Payne, Langhhunn, and Crum (1980), Payne, Langhhunn, and Crum (1981), Slovic, Fischhoff, Lichtenstein, Corrigan, and Combs (1977), Swalm (1966), and Williams (1966). Theorem 3 and t h e remarks in subsection 3.1 imply t h a t t h e conflict often alludedto
in t h e l i t e r a t u r e between averse-prone choice behavior and t h e r u l e s of rationality is nota
conflict with t h e expected utility conditions, e.g., independence, but i s a conflict with a consistency requirement. I t i s possible, f o r example, f o ra
person's p r e f e r e n c e sLk to
b e a n averse-prone r i s k attitude andto
satisfy expected utility conditions f o r each c u r r e n tasset
positionc
and y e t notto
b e framing consistent.This identification of
a
specific normative principle t h a t is in conflict with averse-prone choice behavior suggests a question f o r f u t u r e empirical work. For which persons and in which contexts isa
person's averse-prone choice behavior a r e s u l t ofa
p r e f e r e n c e issue t h a t i s of genuine importanceto
t h e person r a t h e r than a r e s u l t of his information processing limitations and of his limited experi- ence in risk taking. Dickson (1981) found t h a t a risk manager, i.e.,a
person "con- fronted continually with problems holding out t h e chance of loss" i s likelyto
b econsiderably
m o r e
r i s k a v e r s e f o r potential loss-producing choices than i s a non- r i s k manager, i.e.,a
person who "will r a r e l y be concerned with decisions where only a lossor
b r e a k even point i s in prospect". A survey by Freifelder and Smith (1984) h a s similar results. Hershey and Schoemaker (1980a) and Schoemaker and Kunreuther (1979) found t h a t risk aversion is f a r g r e a t e r when decisionsare
presented in a n insurance context than when t h es a m e
decisionsare
presented as standard gambles. Experimental studies in Hershey and Schoemaker (1980b) t h a t examine t h e reflection hypothesis of Kahneman and Tversky (1979)at
both t h e across-sub ject and within-subject levels "seriously question t h e generality of prospect theory's reflection hypothesis."One approach
to
t h e question posed above would b eto
investigate t h e e x t e n tto
which t h esame
person in different contexts and different persons in t h e same context would modify t h e i r averse-prone choice behavior a f t e r being informed asto
i t s conflict with t h e normative principle of framing consistency. Slovic and Tversky (1974) examine similar questions concerning a person's modification of his p r e f e r e n c e s when informed as t o i t s conflict with Savage's independence principle for expected utility.4. C o m b i n a t i o n s o f L o t t e r i e s
Tversky and Kahnemm (1981), (1986) discuss a type of dominance principle t h a t i s concerned with t h e sum of two lotteries. In t h i s section, a g e n e r a l defini- tion of t h i s principle i s specified, and t h e principle i s shown
to
b e violated by any averse-prone r i s k attitude. Dominance concerning sums and multiples of l o t t e r i e s a p p e a r s t o b e a p p r o p r i a t e asa
normative principle; however, i t i s shownto
have implications t h a t r e n d e r i t i t f a r too r e s t r i c t i v e . Indeed, in a n expected utility model, t h i s t y p e of dominance i s shownto
imply t h a t t h e decision maker i s r i s k neu- tral.Consider t h e specific choice problem due t o Tversky and Kahneman (1981) t h a t i s described in Section 2. While most of t h e students in t h e sample (73 per- c e n t ) p r e f e r r e d A t o B and also p r e f e r r e d D
to C ,
none of t h e s e students p r e - f e r r e d t h e sum of A and Dto
t h e sum of B and C:Problem 2. Choose between:
A and D. 25% c h m c e t o win 8240, and
75% c h a n c e
to
lose 8760. [0 p e r c e n t ] B andC.
25% c h a n c e t o win 8250, and75% chance
to
lose 8750. [ l O O p e r c e n t ]Tversky and Kahneman a r g u e t h a t t h e students' decisions
are
"violations of t h e r u l e s of rational choice" and, more specifically, "violations of dominance" f o r t h o s e students who p r e f e r r e d Ato B
and p r e f e r r e d Dto
C even though t h e sum of B andC
stochastically dominates t h e sum of A and D.I t will b e useful t o define precisely t h e principle t h a t i s being violated by t h e students' decisions in t h i s example. Definition 4 below i s intended
to
d o so; t h e r e are o t h e r definitions t h a t havea
g r e a t e ror
lesser a p p a r e n t generality butare
in f a c t equivalent.D e f i n i t i o n 4. A p r e f e r e n c e relation on l o t t e r i e s will b e called
sum-dominance c o n s i s t e n t
provided t h a t f o r m y probability 0<
p<
1, both of t h e p r e f e r e n c e sd o not o c c u r whenever t h e l o t t e r y
<
p S(z+
z; ), Q S(z+
z j )>
i s dominated by t h e l o t t e r y < p S ( z 2+
z; ), qS(z3 + Z i )>
in t h a tIn t h e above Problems 1 and 2, p
=
.25 and zl=
8240,z2 =
$1000, 2 3=
80, Zi=
-$750, 2;=
SO, a n d z j = -8l000.
Before discussing t h e implications of sum-dominance consistency, w e will describe a second, analogous type of dominance. To focus t h e discussion, consider
a
person who wishes t o investa
c e r t a i n amount of money, which will be r e f e r r e dto
as t h e person'scurrent
f i n d . I t is meaningful t o measure t h e consequences of t h e available investment l o t t e r i e s not only by net gains and losses but also by percent increases and d e c r e a s e s in t h e person's c u r r e n t fund. Consider, f o r example, t h e following choice problem.Problem 3. Imagine t h a t you f a c e t h e following p a i r of c o n c u r r e n t decisions.
First examine both decisions, then indicate t h e options you p r e f e r .
Decision (i). Choose between:
A.
a
s u r e gain of 6XB.
a
one-fourth chance t o gain 25X. anda
three-fourths chance t o gain nothingDecision (ii). Choose between:
C. a s u r e loss of 15X
D.
a
three-fourths chanceto
lose 20%. anda
one-fourth chanceto
lose nothing.This problem h a s not been empirically tested. However, i t i s similar t o t h e Problem 1 t e s t e d by Tversky and Kahneman in t h a t mild risk aversion f o r gains leads t o a p r e f e r e n c e of A o v e r B and any d e g r e e of r i s k proneness f o r losses leads
to
a p r e f e r e n c e of D o v e r C.Now, consider t h e sequential o c c u r r e n c e of t h e investment lotteries in t h e de- cisions (i) and (ii). (The resulting p e r c e n t changes
are
t h esame
r e g a r d l e s s of whether decision (i) o r decision (ii) o c c u r s first.)Problem 4. Choose between:
A and D. a one-fourth chance
to
gain 6X a three-fourths chanceto
lose 15.2X B and C. a one-fourth chanceto
gain 6.25Xa three-fourths chance
to
lose 15XNote t h a t t h e combination of t h e lotteries B and C dominates t h e combination of t h e l o t t e r i e s A and D in this problem just
as
t h e combination of B and C dom- inates t h e combination of A and D in Problem 2.In general,
let z
denote a p e r c e n t increase ( z >OX), a p e r c e n t d e c r e a s e (-100%< z <
OX), or no change ( z=
OX) ina
person's c u r r e n t fund. Let<
p (zX), q (z'X)>
denotea
lottery havinga
p e r c e n t gain or loss ofz
with proba-bility p and a p e r c e n t gain or loss of
z
' with probability q=
1-
p.
Note t h a t if f i r s t a p e r c e n t changez l
o c c u r s and seconda
p e r c e n t change z 2 occurs, then t h e overall p e r c e n t change isz + z z +
( z l z 2/ 100). Letzl0z2
denote this resulting p e r c e n t change.Definition
5.
A p r e f e r e n c e relation on l o t t e r i e s will be called percent- dominance c o n s i s t e n t provided t h a t f o r any probability 0<
p<
1 , both of t h e p r e f e r e n c e sd o not o c c u r whenever t h e l o t t e r y <p(z1oz; X), q (zlOz j X)
>
is dominated by t h e lottery <p ( z 2 0 z i X), q(z30zi X)>
in t h a tZ ~ O Z ~
<
Z ~ O Z ~ andzlozj <
2302;.
(12)In t h e above Problems 3 and 4, p
=
.25 andz l =
6X,z 2 =
25X,z 3 =
OX,Z i
=
-15%. Z i=
OX, andz j =
-20%.Definitions 4 and 5 formalize c e r t a i n requirements on p r e f e r e n c e s
as
pro- posed normative principles. The implications of t h e s e proposed principlesare as
follows.Theorem 4. Consider a decision problem in which t h e possible consequences are financial changes measured e i t h e r as absolute gains and losses
z
o r as p e r c e n t gains and lossesz .
Assume t h a t t h e decision maker's p r e f e r e n c e relation on lot- t e r i e s satisfies (i) t h e conditions of expected utility and (ii) t h e condition t h a t l a r g e r amountsare
p r e f e r r e d . Then:(a) The p r e f e r e n c e relation is sum-dominance consistent if and only if t h e de- cision maker h a s
a
constant risk attitude f o r possible net gains and losses.(b) The p r e f e r e n c e relation is percent-dominance consistent if and only if t h e decision maker h a s a constant proportional risk attitude f o r possible amounts of his c u r r e n t fund.
(c) Therefore, t h e p r e f e r e n c e relation i s both sum-dominance consistent and percent-dominance consistent if and only if t h e decision maker is risk neutral.
Sum-dominance consistency and percent-dominance consistency a p p e a r
to
b e reasonableas
normative principles. They exclude t h e type of choice behavior t h a tw a s
observed by Tversky and Kahneman (1981) f o r Problems 1 , 2, and t h a t is con- jecturedto
o c c u r f o r Problems 3, 4. A s Theorem 4 demonstrates, however, these principles exclude not only a n averse-prone risk attitude but any attitude toward risk o t h e r than r i s k neutrality.The assumptions (i), (ii) in Theorem 4 r e q u i r e t h a t t h e r e exists
a
s t r i c t l y in- creasing utility function f o r t h e p r e f e r e n c e relation. They d o not require, howev- e r , t h a t the utility function h a s derivatives of any o r d e ror
eventhat
i t is continu- ous. Thus, Theorem 4 applies in p a r t i c u l a rto
t h e p r e f e r e n c e relation f o r t h e ef- f e c t s context in a reward-penalty model.5. Uses of the Reward-Penalty Yodel
This section discusses f i r s t t h e possible testing of t h e reward-penalty model
as a
descriptive model, and second i t s possible usefulness as a prescriptive model in a decision analysis study.The reward-penalty model is sufficiently r e s t r i c t i v e
to
b e testableas a
descriptive model of p r e f e r e n c e s in choice behavior. In this sense, i t is not sub- jectto
t h e criticism of Tversky (1977) t h a t introducing additional variablesas
descriptors of psychological consequences i s a d hoc. However, any empirical study of the model t h a t comparesa
person's p r e f e r e n c e s int w o
different contexts (whatw e
have oalled t h e e f f e c t s context and t h e no-effects context) will needto
b e carefully designed e i t h e rto
conveyto
t h e persons being interviewed aclear
understanding of t h e e f f e c t s context and of t h e no-effects contextor to
elicit by indirect questioning t h e i r p r e f e r e n c e s in t h e no-effects context. The design prob- lems will b e simpler f o r a n empirical study t h a t considers only t h e effects context andtests
t h e implications (a)-(f) in Theorem 2 of a reward-penalty model. Sucha
study could o f f e r indirect evidenceas to
t h e descriptive accuracy of a reward- penalty model f o r different types of persons and under different circumstances.The evidence is strong t h a t
w e
humans have systematic and stubborn weaknesses in o u r abilityto
process information onour
p r e f e r e n c e s and o u r pro- bibalistic beliefs (see, f o r example, Kahneman, Slovic, and Tversky, 1982 and t h e review by Schoemaker, 1982). For this very reason, t h e distinction between descriptive models and prescriptive models can b e useful; descriptive models have t h e purpose of aidingour
understandingas
scientists of how people make decisions and prescriptive models have t h e purpose of aiding o u r ability as decision makersto
process information f o r making decisions (see, f o r example, Bell, Raiffa, and Tversky, 1984, Howard, 1980, Keeney, 1982, and Raiffa, 1961.) Leaving aside t h e operational difficulties of assessing simple preferences, this argument depends on t h e premise t h a t a p r e s c r i p t i v e model can c a p t u r e t h e major p r e f e r e n c e concerns of t h e decision maker. Thus,it
is idso importantto
distinguish between those behavioral deviations from a p r e s c r i p t i v e model t h a tare
dueto
suboptimal infor- mation processing, i.e., heuristic biases, on t h e p a r t of t h e decision maker and those deviations t h a t are dueto
p r e f e r e n c e concerns of t h e decision maker t h a tare
not included in t h e model.The reward-penalty model i s prescriptive r a t h e r t h a n descriptive in t h a t i t is not intended
to
modela
person's heuristic biases. R a t h e r , i t i s intendedto
identify those p s y c h o l o g i d e f f e c t s thatare
r e g a r d e d by the decision makeras
a n impor- t a n t a s p e c t of his consequences (and t h a t cause his p r e f e r e n c e sto
deviate from t h e p r e f e r e n c e condition of framing consistency). The recognition of t h e s e effects in t h e model isto
b e r e g a r d e das
t h e inclusion ofa
p r e f e r e n c e issue of importance to t h e decision maker r a t h e r thanas
t h e inclusion of systematic imperfections in t h e reasoning of t h e decision maker. Thus, t h e purpose of t h e reward-penalty model ism o r e
r e s t r i c t e d than t h e purposes of such general models as p r o s p e c t theory (Kahneman and Tversky, 1979), weighted utility theory (Chew and MacCrim- mon, 1984), and SSB theory (Fishburn, 1982).In addition
to
t h e question of whether averse-prone choice behavior may ac- curately r e f l e c t a decision maker's p r e f e r e n c e s , and in t h a t sense is rational, t h e r e is also t h e question of whether such choice behavior i s in t h e b e s t i n t e r e s t of a n organizationor
of society, and in t h a t sense is ethical. The psychological qualities t h a tare
hypothesized in Section 1to
b e associated with averse-prone choice behavior are not flattering. In t h e case of a n individual who is acting on his own behalf, t h e s e qualities can b e negatively describedas
ego-centric and inflexi- ble; t h e person would r a t h e r r i s k losing a l a r g e r amount than admit evento
himself t h a t h e h a s suffered a loss. In t h e case of a n individual who is acting on behalf ofa n organization, t h e r e is t h e additional quality t h a t his concern f o r his own repu- tation may
at
times b eat
variance with concerns f o r t h e effectiveness of t h eor-
ganization; in common parlance, t h e person may be primarily concerned with "cov- ering his ass."Thus, f o r decision analysis applications in which a n averse-prone r i s k attitude is a n important f e a t u r e of t h e decision maker's p r e f e r e n c e s , t h e r e may b e cogent reasons f o r formulating
t w o
versions of t h e decision analysis model, one t h a t in- cludesn
and pas
additional parameters. and one t h a t does not (and thus assumes t h a t p r e f e r e n c e s are framing consistent). Both t h e effects context and t h e no- effects context would t h e r e b y by considered.This caution
as
r e g a r d s t h e modeling of a decision maker's complete p r e f e r - e n c e s is similarto
t h e point made in Bell (1985)"that
what i s c u r r e n t l y omitted from expected utility analysis deservesto
b e omitted and t h a t a formal analysis may b e exactly what is neededto
prevent a decision maker's intuition from forcing economically inefficient decisions [italics Bell's]."
This point is madealso
in Raiffa (1984). In a decision analysis application f o r a n organization, t h e most important p r e f e r e n c e sto
b e examined may b e those of a n exemplifying individual r e p r e s e n t - ing t h e e n t i r e organization r a t h e r than those of any single individual within t h eor-
ganization.Appendix: Proofs of resulta
Proof of Theorem 1. If u ( z ) is any utility function f o r t h e no-effects context, then
u
( z )=
U(z , z l*,..., zn
* ) f o r some utility function Uas
described in (2). If w ( z ) is any utility function f o r t h e e f f e c t s context, then w ( z )=
f i ( z , r l(z),... , r n ( 2 ) ) f o r some (possibly different) utility function
fi as
in (2).hforeover,
fi =
a U+
b f o r some constants a>
0 and b . Therefore,Proof of
Theorem
2. Assume t h a t t h e utility functionu
f o r t h e no-effects context is normalized so t h a tu
( 0 )=
0. Let L denote t h elottery to
b e compared.To prove (a), observe t h a t f o r
z <
0 , w ( p z )=
u ( p z-
n)=
u(-n)+
o ( 1 ) while w ( 1 )=
p w ( z )+
q w ( 0 )=
p u ( z - n )=
p u ( - n )+
o ( 1 ) . Thus p u ( ~ r )> u(--rr)
implies w ( 1 )>
w ( p z ) f o rz <
0 sufficiently n e a rto
0.To p r o v e (b), o b s e r v e t h a t f o r
z >
0 , w ( p z )=
u ( p z+
p) while w ( 1 )=
PU ( 2 + P )
+
q u ( 0 ) . Sinceu
is strictly concave and p r 0 ,+ p) + q u ( O )
<
u ( p ( z + p ) ) S u ( p z + p ) , a n d t h u s w ( 1 )<
w ( p z ) . To prove (c), i t sufficesto
considerz >
0 . Then,since p
<
n and u is s t r i c t l y increasing.To p r o v e (d),
w e
must show t h a t w ( 1 ) = p u ( z-
n) is less than w ( p z )= u
( p z -n) f o r sufficiently l a r g e negativez .
A sa
f i r s t observation, t h e condition of unbounded utility from below is well-knownto
imply t h a t t h e strictly increasing function U ( Z ) is unbounded from below (hence t h e name). There aret w o cases
t o consider: f i r s t , t h a t in which t h e amountsz
havea
finite lower bound a (with a<
-n so t h a t w( z )
i s defined f o r somez <
0 ) and limu ( z ) = -,
and= + a +
second, t h a t in which t h e amounts
z
have no lower bound and lirn u( z ) = -.
r -.-
In t h e f i r s t case, r -(a +n)+ l i m p u
( z
-7r)= -
while r +(a +n)+ lim u ( p z -n)=
u
( a +(p -l)(a + n ) ) is finite since (p -l)(a +n)>
0. Therefore, p u( z
--T)< u
(pz -n) f o r allz
less than some amountzp
between a+
n and 0.In t h e second c a s e , t h e condition of decreasing risk aversion implies t h a t lirn
u ' ( z ) = +-.
To show this, note t h a tu
" ( 2 ) 6 0 f o r allz
impliesthat u ' ( z )
is r -.-decreasing f o r all
z .
Thus, lirnu ' ( z )
isa
finite number b o r is +a. Ifr
--
lirn
u ' ( z ) =
b,
then lim s u p ~ " ( 2 )=
0 and hence lirn inf -u" ( z ) / u
' ( 2 )=
0.r
--
r +- r--However, decreasing risk aversion implies t h a t t h e local risk aversion function -u " ( 2 )/
u
' ( 2 ) i s positive and decreasing ( P r a t t , 1964). and this p r o p e r t y contra- dicts t h e previous statement.Now, assume t h a t 0
<
p<
1 i s fixed. If n=
0 , then f o r anyz <
0 t h e line from t h e point( z , u
( 2 ) )to
t h e point (0,O) lies below the graph ofu ,
and t h e r e f o r e p u ( z )+
q u ( 0 ) = p u ( z )<
u ( p z+
9.0)=
u ( p z ) .If n
>
0 , then t h e r e exists a numberz o < -
n such t h a t t h e line from(zo,u ( z o ) )
t o (-n.0) lies above t h e g r a p h ofu .
Since lirn r +- u ' ( 2 )= -
andu ( z )
i sunbounded from below, t h e r e exists a unique
z l < z o
such t h a t this line intersectst h e g r a p h of u
alsoat ( z l , u (xi)). Moreover,
forany zp < z l , t h e line
f r o m( z p , u ( z p ) ) to (-n,O)
w i l l alsoi n t e r s e c t t h e graph of u at a unique point (z; , u (z;
))with z0 <
Z;< -
TT.Choosing zp so t h a t pzp -
TT< z; ,
itfollows t h a t f o r any amount z < zp , pu ( z - TT) +
q- 0 < u
(p(Z -TT) +
q(-TT)) and hence
pu(Z -TT) < u (pz -TT)
as w a sto
beshown.
To p r o v e (e), note t h a t t h e certainty equivalent
of t iszo = u
-'(pa( z ) +
qu( z
'))in t h e no-effects context and zl = w -l(pw ( z ) +
qw( z
'))where w ( t
)= u ( t -TT)
ist h e
effectscontext. Since u r e p r e s e n t s a n attitude of de- creasing r i s k aversion,
it followsby
ar e s u l t in P r a t t (1964, Theorem
2)t h a t
Z1
<
Zo.Finally, t h e
proof ofu )
is similarto t h a t
of (e)above e x c e p t t h a t w ( t ) = u ( t + p), p > 0, and hence zl > zo.
Proof of
Theorem
3.Suppose t h a t f o r t w o different c u r r e n t asset positions
c
<
c',t h e p r e f e r e n c e relations kk and satisfy t h e condition
ofa n averse-
p r o n e risk attitude. Then, framing consistency implies t h a t t h e p r e f e r e n c e
rela-tion &
isr i s k a v e r s e on t h e r a n g e
y>
cand r i s k prone on t h e r a n g e
y<
c ' .For t h e intersection range, c <
y< c
',t h e s e r i s k attitudes are contradictory.
Proof of
Theorem
4. To show part(a), w e
w i l l relatet h e condition of sum- dominance consistency to t h e condition of a constant r i s k attitude. Let I denote t h e interval
of possiblemonetary changes z. One definition H m e y (1981, 1986)
of aconstant risk attitude
ist h a t
forany amounts
h l<
h 2<
h 3and any probability 0
< p< I , if
f o r some z such t h a t z + h i , z + h o , and z +
h 3are in I, then
f o r a n y z ' s u c h t h a t z ' +
hl.z' + h 2 , a n d z ' + h 3 a r e i n I .
For
ap r e f e r e n c e relation 2 t h a t satisfies t h e conditions
ofexpected utility
and such t h a t l a r g e r amounts z
arep r e f e r r e d , t h i s condition holds if and only if 2
c a n
ber e p r e s e n t e d by a utility function of t h e form
f o r some parameter value
r
(Harvey, 1986). Note t h a t no assumptions of differen- tiability o r even of continuity of t h e utility functionu
( z ) are required.Assume t h a t t h e r e is a constant risk attitude and, in particular, t h a t ;G is represented by a utility function u ( z )
=
e x p ( r z ) f o r somer >
0. Consider amountszi
, Z i i=
1,2,3, and a probability 0<
p<
1 as in Definition 4. Then,Therefore, t h e preferences (9) imply t h a t
which implies t h a t (10) i s false. Thus,
2
is sum-dominance consistent. Similar ar- guments can b e given f o r t h ecases u
( 2 )= z
andu
( z )=
--exp(rz ),r <
0.To show t h e converse implication, f i r s t assume t h a t t h e r e i s not a constant risk attitude but t h a t t h e utility function
u
( z ) i s continuous. Then, t h e r e exists amountsz +
h i ,z ' +
h i , i=
1,2,3, in I and a probability 0<
p<
1 such t h a t (Al) is t r u e but (A2) i s false. Bya
slight change in p , i t follows t h a t t h e indifferences in (Al), (AZ) can b e replaced by opposite preference. Without loss of generality, assume t h a t t h e r e i s t h e p r e f e r e n c e s+
in (Al) and t h e preference 4 in (A2).Then, by t h e continuity of u ( z ) , these preferences will remain t r u e when
z +
h l in (Al) is changedto a
slightly smaller amountz +
h i-
d , d>
0. I t follows t h a t f o r~ ~ = ~ + h ~ - d , ~ 2 = ~ + h ~ , z 3 = z + h 3 and Z i = Z ' + h l , Z i = z f + h 2 ,
z; = z' +
h 3 t h e conditions (9) and (10) are satisfied, i.e., t h e r e isa
violation of sum-dominance consistency.Second, assume t h a t t h e utility function u ( z ) is not
a
continuous function on t h e interval I. Sinceu
( z ) i s strictly increasing, i t has only a finiteor
countably infinite number of points of discontinuity. Thus,w e
may choose a pointz l at
whichu
( z ) is discontinuous anda
pointZ i at
whichu
( z ) is continuous. Suppose, f o r ex- ample, t h a t sup{u
( 2 ) :z < z l j < u
(zl). Choose h>
h '>
0 such t h a tz 2 = z 1 +
h a n d z i= z i +
h ' a r e i n 1 . There e x i s t s a p r o b a b i l i t y 0 < p < l s u c h t h a tf o r all
z < z
Sinceu
( z ) is continuousat z i
, w e can choose anz j < z i
suff i- ciently n e a rto z i so
t h a tNow, choose
z3
sufficiently n e a r t ozl
s o t h a tz l -
2 3 isless
thanZ i - z$ .
T h e n , z l + ( z i + h f ) < ( z l + h ) + z i a n d z l + z j < z 3 + z i . H e n c e , t h e r e i s a violation of sum-dominance consistency. A similar argument can be given f o r t h e case i n f i u ( z ) :
z
> z l ]>
u ( z l ) .W e will derive p a r t (b) of Theorem 4 from p a r t (a) by means of a change of variable argument. Suppose that t h e decision maker's c u r r e n t fund is denoted by a variable y
=
c+ z
where c is his initial fund andz
is t h e subsequent net change in t h e fund. W e assume t h a t y>
0 f o r all net gains o r lossesz
in t h e interval I.Consider a new variable, w
=
log y.
The preference relation associated with y , which w e will denote h e r e byh e
induces a preference relation associated with w , whichw e
will denote by&,.
I t is well-known that t h e preference relation
;L21
satisfies t h e condition of a constant proportional risk attitude if and only if t h e preference relation&
satis- fies the condition of a aonstant risk attitude (see, e.g., Harvey, 1986 f o r a detailed discussion). Thus, t o prove p a r t (b), i t sufficesto
show t h a t is percent- dominance consistent if and only if &, is sum-dominance consistent.A lottery
<
p (z X), q (z 'X)>
expressed in terms of percent changes z,
z ' is equivalent t o t h e lottery<
p S b,
q Sk 'y>
with k=
1+
(z / l o o ) , k '=
1+
(z '/ 100) expressed in terms of asset positions. Therefore. Definition 5 can be restated as t h e condition t h a t f o r any probability 0<
p<
1 both of t h e preferencesdo not occur whenever
klk;
<
k 2 k i and k l k j < k 3 k i.
(A41However. a lottery
<
p *ky,
q *k 'y>
f o r t h e preference relation&
corresponds t o a lottery<
p (Log k+
Log y ), q (Log k '+
Log y )>
f o r t h e preference relation&,.
Thus, t h e r e exist amounts ki , k i,
i=
1.2.3. such t h a t (A3), (A4)are
satisfied if and only if t h e r e exist amounts z i ,Zi
, i=
1.2,3, such that (9), (10)are satis-
fied.P a r t (c) follows immediately from p a r t s (a) and (b) since a utility function