137
N O N E X P E C T E D - U T I L I T Y T H E O R Y
the inverted-s with the consequent overweighting of “small” probabilities below
the inflection point, and underweighting above it. With •
,
1, • (
?
) is “sub-certain”
in the sense that the sum of weights (•
i
• (
p
i
)) will be less than unity. Lattimore,
Baker, and Witte (1992, p. 381) describe this as “ ‘prospect pessimism’ in the sense
that the value of the prospect is reduced vis-à-vis certain outcomes.” In their em-
pirical estimates, they find that allowing nonlinear decision-weights offers signifi-
cant improvement in predictive power over EU (which is the best model for only
about 20 percent of their subjects). The best-fitting weighting function is generally
the inverted-s exhibiting greater sensitivity to high and low probabilities relative to
mid-range probabilities. They also report differences between the best-fitting
weighting functions for gains and losses (for example “pessimism” is more pro-
nounced for losses), though the interpretation of these differences is potentially
confounded by the fact that, in their study, gains are measured in units of money
while losses are measured in units of time.
Single-parameter weighting functions have been proposed by Tversky and
Kahneman (1992) and Prelec (1998). Tversky and Kahneman suggest the form •
(
p
)
5
p
/[(
p
1
(1
2
p
)
•
)
1/•
]. This generates the inverted-s for 0
,
•
,
1, and re-
ducing • lowers the crossover point while accentuating the curvature of the func-
tion. Their empirical analysis supports the s-shaped weighting function and also
reveals systematic differences in behavior for gains and losses: specifically, indif-
ference curves in the best-fitting models for losses resemble those for gains
flipped around a 45 degree line. This supports the case for a model that distin-
guishes between gains and losses (i.e., a model with a reference point), though
virtually no work is done by the weighting function here; essentially, the same
probability-weighting function works well for both gains and losses.
Prelec proposes the function • (
p
)
5
exp(
2
(
2
ln p
)
•
). With 0
,
•
,
1, this gen-
erates the inverted-s with a fixed inflection point at
p
5
1/
e
5
0.37. Visually, • is
the slope of • (
?
) at the inflection point, and as • approaches unity, • (
?
) becomes ap-
proximately linear; as it approaches zero, • (
?
) approximates a step function. Prelec
argues that a crossover in the vicinity of 1/
e
is consistent with the data observed
across a range of studies. A novel feature of Prelec’s contribution is to provide an
axiomatization for this form, and he also discusses a two-parameter generalization.
The two-parameter version is similar in spirit to the “linear in log odds form” pro-
posed by Gonzalez and Wu (1999) in that it allows the curvature and elevation of
the weighting function to be manipulated (more or less) independently. In the lat-
ter form, probability weights are given by
(10)
The parameter • primarily controls the absolute value of • (
?
) by altering the ele-
vation of the function, relative to the 45-degree line, while • primarily controls
curvature. Gonzalez and Wu’s data suggests that the flexibility of a two-parameter
model may be useful for explaining differences among individuals. For other pur-
poses, however, parsimony favors the one-parameter versions.
•
•
•
=
+
−
•
•
•
( )
/ [
(
) ].
p
p
p
p
i
i
j
i
1
Do'stlaringiz bilan baham: |