125
N O N E X P E C T E D - U T I L I T Y T H E O R Y
function (see section 5.1.1). A useful discussion of the theoretical properties nec-
essary and sufficient for an s-shaped weighting function can be found in Tversky
and Wakker (1995).
Axiomatizations of rank-dependent expected utility have been presented by,
among others, Segal (1990), Wakker (1994), Abdellaoui (1999), and Yaari (1987),
who examine the special case of the model with linear utility (this is essentially a
rank-dependent reformulation of Handa’s proposal with
u
(
x
i
)
5
x
i
. Wakker, Erev,
and Weber (1994) provide a useful discussion of the axiomatic foundations of
rank-dependent expected utility in which they demonstrate the essential differ-
ence between EU and rank-dependent expected utility is that the latter theory re-
lies on a weakened form of independence called “comonotonic independence.” It
is an implication of the standard independence axiom that if two prospects
q
and
r
have a common outcome
x
, which occurs with probability
p
, in each prospect,
substituting
x
for some other outcome
y
in both prospects will not affect the pref-
erence order of
q
and
r
. The same may not be true in the rank-dependent model,
however, because such substitutions may affect the rankings of consequences and
hence the decision weights. Comonotonic independence asserts that preferences
between prospects will be unaffected by substitution of common consequences so
long as these substitutions have no effect on the rank order of the outcomes in
either prospect.
Various generalizations of the rank-dependent model have been proposed
(Segal 1989, 1993; Chew and Epstein 1989; Green and Jullien 1988). In Green
and Jullien, the crucial axiom is ordinal independence. Suppose two prospects
q
,
r
have a “common tail” such that for some
j
,
p
qi
5
p
ri
for all
i
from
j
to
n
. Ordinal
independence requires that preferences between
q
and
r
be unaffected by the sub-
stitution of this common tail, in both prospects, with any other common tail. This
axiom is necessary for any rank-dependent model. The contribution of Chew and
Epstein constructs a theoretical bridge between the rank-dependent models and
the betweenness-conforming theories (i.e., those with linear indifference curves
discussed previously) by presenting a general model that contains each class as a
special case (see also the “correction and comment” by Chew et al. 1993).
A further extension to the rank-dependent model discussed by Starmer and
Sugden (1989), Tversky and Kahneman (1992), and Luce and Fishburn (1991)
involves a distinction between consequences that are “gains” and those that are
“losses.” This approach draws on Kahneman and Tversky’s earlier work on
prospect theory. It is to this model that we now turn, and in doing so we cross the
boundary into nonconventional territory.
4.2. Nonconventional Theories
4.2.1. THE PROCEDURAL APPROACH AND REFERENCE DEPENDENCE
Each of the theories we have considered so far models choice as preference maxi-
mization and assumes that agents behave
as if
optimizing some underlying prefer-
ence function. The “as if” is significant here: the conventional approach, interpreted
Do'stlaringiz bilan baham: |