132
S T A R M E R
preference altogether. Can we speak of people maximizing anything if they don’t
have transitive preferences? It turns out that the answer is yes.
There is at least one well-known theory of choice based on a model of nontran-
sitive preference. The theory I have in mind was proposed simultaneously by Bell
(1982), Fishburn (1982), and Loomes and Sugden (1982). I will begin by dis-
cussing a version of this theory presented by Loomes and Sugden (1987), which
they call
regret theory
. Its central premise is closely akin to the psychological in-
tuition at the heart of the theory of disappointment. In that theory, it is assumed
that an individual compares the outcomes
within a given prospect
giving rise to
the possibility of disappointment when the outcome of a gamble compares unfa-
vorably with what they might have had. Regret theory allows comparisons among
consequences to affect choice, but in this case, the relevant comparisons occur
among the consequences of alternative choice options.
Since the theory has to allow comparisons among choice options, it cannot be a
conventional theory that assigns values independently to individual prospects.
Loomes and Sugden propose a theory of pairwise choice in which preferences are
defined over pairs of
acts
, where an act maps from
states of the world
to conse-
quences.
21
Let
Ai
and
Aj
be two potential acts that result in outcomes
x
is
and
x
js
,
respectively, in state of the world
S
. The utility of consequence
x
is
is given by a func-
tion
M
(
x
is
,
x
js
) which is increasing in its first argument and decreasing in its second.
This function allows the utility from having
x
is
be suppressed by “regret” when
x
is
,
x
js
, or enhanced by “rejoicing” when
x
is
.
x
js
. The individual then seeks to maxi-
mize the expectation of modified utility í
s
p
s
?
M
(
x
is
,
x
js
) where
p
s
is the probability of
state
S
. Regret theory reduces to EU in the special case where
M
(
x
is
,
x
js
)
5
u
(
x
is
).
Although preferences are defined over acts, the theory can be applied to
choices among prospects given some assumption about how outcomes are corre-
lated between them. One interesting case is when consequences are uncorrelated
between prospects; that is, when prospects are
statistically independent
. In a
choice between a pair of such prospects
q
and
r
,
if q
is chosen, the probability of
getting
x
i
and missing out on
x
j
is given by
p
qi
p
rj
where
p
qi
is the probability of
consequence
x
i
in
q
and
p
rj
the probability of
x
j
in prospect
r
. Preferences between
q
and
r
are then determined by the expression
(7)
where • (
x
is
,
x
js
) •
M
(
x
is
,
x
js
)
2
M
(
x
js
,
x
is
). The function • (
?
,
?
) is skew symmetric
by construction, hence • (
x
,
y
) •
2
• (
y
,
x
) and •(
x
,
x
) • 0 for all
x
,
y
.
If prospects are statistically independent, the addition of a further assumption,
which Loomes and Sugden call
regrets-aversion
,
22
implies that indifference curves
q
r
p p
x x
i
j
qi
ij
i
j
~
( , )
,
a
s
⇔
=
<
>
• •
•
0
21
As a theory of pairwise choice, regret theory has limited applicability, but ways of generalizing
the theory have been suggested by Sugden (1993) and Quiggin (1994).
22
In their early discussions of regret theory, Loomes and Sugden called this assumption “convexity.”
Do'stlaringiz bilan baham: |