Advances in Prospect Theory: Cumulative Representation of Uncertainty Amos Tversky Stanford University, Dept. of Psychology Daniel Kahneman U. C.

Berkeley, Dept. of Psychology A new version of prospect theory: Cumulative Prospect Theory Applies to both uncertain and risky prospects New experiment confirms fourfold pattern of risk aversion: i. Risk-aversion for gains of high probability ii. Risk-seeking for losses of high probability iii. Risk-seeking for gains of low probability iv. Risk-aversion for losses of low probability Expected utility theory has come under serious question in the 1980s. Prospect Theory explains major violations of Expected Utility Theory with a small number of outcomes.

Prospect Theory: 1. A value function that is concave for gains, convex for losses, and steeper for losses than for gains. 2. Nonlinear transformation of the probability scale, which over-weights small probabilities, and which under-weights moderate and high probabilities. New Representation: Cumulative Functional, which transforms cumulative rather than individual probabilities. Cumulative Prospect Theory unifies the Prospect Theory and the Cumulative Functional representation.

Setting the stage for present development: Five phenomena: 1. Framing Effects: Equivalent formulations of a choice problem do not necessarily give rise to the same preference order. (Unlike Arrow, 1982) 2. Non-linear Preferences: Non-linear preferences even in choices that do not involve sure things (Allais, 1953) 3. Source Dependence: Peoples willingness to bet on an uncertain event depends not only on the degree of uncertainty but also on its source. 4. Risk seeking: Unlike the traditional risk aversion assumption in economic analyses of decision under uncertainty. 5. Loss aversion: Losses appear to be larger than gains. The asymmetry is too extreme to be explained by income effects or by decreasing risk-aversion. 1. Theory: Prospect Theory says that choice process consists of two phases: framing and valuation. The valuation process discussed in subsequent analyses is applied to framed prospects. 1.1. Cumulative Prospect Theory S : Finite set of states

Subsets of S are called events. X: Set of consequences/outcomes. Neutral outcome, 0, gains, > 0, and losses, < 0. f : S X is an uncertain prospect. To define cumulative functional, arrange the outcomes in increasing order. A prospect is then represented as {( xi , Ai )}, which yields xi if Ai occurs, where xi > xj iff i > j, and (Ai) is a partition of S. f +(s) = f (s) if f (s) > 0, and f +(s) = 0 if f (s) <= 0. Utility function: V(f) > = V(g) iff f > = g. Capacity function: for A S, this function assigns a number W(A) satisfying W() = 0, W(S) = 1, and W(A) > = W(B) whenever A B. v : X Re with v ( x 0 ) = v (0) = 0 , for f = ( xi , Ai ), -m < = i < = n,

V( f ) = V( f + ) +V( f )
V ( f
+

) =

i=0

+ i

v(xi ) ,

V ( f ) =

i= m

v ( xi )

(1)

+ + where i+ ( f + ) = ( 0 ,......, n ), and i ( f ) = ( 0 ,......, n )

are defined by
+ n = W + ( An ), m = W ( Am )

i+ = W + ( Ai ........ An ) W + ( Ai+1 ........ An ), 0 i n 1 i = W ( Am ........ Ai ) W ( Am ........ Ai1), 1 m i 0

Letting i = i+ if i 0, and i = i if i 0, (1) becomes V ( f ) = iv(xi )

i =m n

(2)
and

Note that

i =0

+ i

=1

i = m

= 1.

If W is additive and hence a probability measure then i is simply the probability of Ai. If f = ( xi , Ai ) is given by a probability distribution p(Ai) = pi , then it can be viewed as a probabilistic or risky prospect (xi, pi). then the decision weights are defined by:
+ n = w + ( pn ), m = w ( pm )

i+ = w+ ( pi + ......+ pn ) w+ ( pi+1 + ......+ pn ),0 i n 1,

where w+ and w- are strictly increasing functions from [0,1] into [0,1] with w+(0) = w-(0), and w+(1) = w-(1). Ex: Roll a dice. If x is even, you receive $x, if x is odd you pay $x. f= (-5, -3, -1; 2, 4, 6), f + = (0,1/2 ; 2,1/6 ; 4,1/6 ; 6,1/6), f - = (-5,1/6; -3,1/6; -1,1/6; 0,1/2). Then, V ( f ) = V ( f + ) +V ( f )

i = w ( pm + ......+ pi ) w ( pm + ......+ pi ),1 m i 0.

= v ( 2)[ w + (1 / 2) w + (1 / 3)] + v ( 4)[ w + (1 / 3) w + (1 / 6)] + v (6)[ w + (1 / 6) w + (0)] + v ( 5)[ w (1 / 6) w (0)] + v ( 3)[ w (1 / 3) w (1 / 6)] + v ( 1)[ w (1 / 2) w (1 / 3)]
1.2. Relation to previous work In original prospect theory: w-(p) = w+(p). In rank-dependent models: w-(p) = 1- w+(1-p), or W-(A) = 1- W+(S-A). Present theory extends the original Prospect Theory. i. Applies to any finite prospect and can be extended to continuous distributions. ii. Applies to probabilistic as well as uncertain prospects, and can, therefore, accommodate some form of source dependence. iii. Allows different decision weights for gains and losses, unlike w- = w+. iv. Satisfies stochastic dominance, and improvement upon the rank-dependent models. 1.3. Values and weights Risk aversion or risk seeking are determined jointly by the value function and the capacities (cumulative weighting functions). Convex v principle of diminishing sensitivity. v steeper for losses than gains principle of loss aversion. Diminishing sensitivity weighting functions, w, concave near 0, and convex near 1. 2. Experiment Goal: to obtain information about v and ws. 25 grad students from Berkeley and Stanford (12 men and 13 women) with no training in decision theory. Each participated in three separate one-hour sessions, and was paid $25. 2.1. Procedure Experiment was conducted on a computer. Computer displayed a prospect and its expected value, along with a descending series of sure outcomes, logarithmically spaced between the extreme outcomes of the prospect. Subject indicated preference between each of the seven sure outcomes and the risky prospect.

Estimate the certainty equivalent in cash. Computer monitored internal consistency of the responses. Present analysis focuses on two-outcome prospects with monetary outcomes and numerical probabilities. A modified procedure also used in eight additional problems. HERE TABLE 3

2.2. Results i. Risk-aversion for gains of moderate and high probability ii. Risk-seeking for losses of moderate and high probability iii. Risk-seeking for gains of low probability provided the outcomes are not extreme. iv. Risk-aversion for losses of low probability provided the outcomes are not extreme. HERE TABLE 4 2.3. Scaling Turn to quantitative description of data. I. Figures 1 and 2 exhibit the characteristic pattern of risk attitudes observed in Table 4. HERE FIGURE 1 and FIGURE 2. Subjects risk-neutral points will lie on diagonal. Subjects risk-averse points will lie below diagonal in figure 1. Subjects risk-seeking points will lie above diagonal in figure 2. Triangles and circles will overlap if the preferences are homogenous. Assuming X = Re, preference homogeneity is both necessary and sufficient to represent v as a two-part power function of the form:
x if x 0 v( x) = ( x ) if x 0

Preference homogeneity appears as a good approximation. Hence, present data can be approximated by a two-part power function. Weighting functions fitted using the functional forms:

w+ ( p) = w ( p) =
II.

p , and ( p + (1 p) )1 p , ( p + (1 p) )1

HERE TABLE 6. For equal chances to win and lose, a prospect will be accepted if the gain is at least twice as large as the loss. Estimation of the model: not to lose generality focus on qualitative aspects rather than quantitative. Non-linear regressions on (5) and (6) imply:

 0.88 , in accord with the principle of diminishing sensitivity.

2.25 , indicating pronounced loss aversion.

Median () = 0.61 and Median () = 0.69. HERE FIGURE 3. Triangle Diagram: HERE FIGURES 4a and 4b. i. Departures from linearity violate expected utility theory. ii. Indifference curves exhibit both fanning in and fanning out. iii. Curves are concave in the upper part of the triangle and convex in the lower right. iv. The indifference curves for non-positive prospects resemble the curves for nonnegative prospects reflected around the 45o line, which represents risk-neutrality. This feature distinguishes the present theory from the rank-dependent model.

Appendix: Axiomatic Analysis Let F = { f : S X } be the set of all prospects, and let F+ and F- denote positive and negative prospects, respectively. A pair of prospects f, g F are comonotonic if there are no s, t S such that f(s) > f(t) and g(t) > g(s). Cumulative prospect theory also satisfies a property called double matching: for all f, g F, if f + g + and f - g -, then f g. For x X, f F, and r S, let x{r}f be the prospect that yields x in state rand coincides with f in all other states. Then a preference relation satisfies tradeoff consistency if for all x, x1, y, y1 X, and f, f1, g, g1 F, and s, t S, we have x{s}f y{s}g, x1{s}f y1{s}g, and x{t}f1 y{t}g1 imply 1 1 x {t}f y1{t}g1. Tradeoff consistency ensures that utility intervals can be consistently ordered. A preference relation satisfies comonotonic tradeoff consistency if tradeoff consistency holds whenever the prospects x{s}f, y{s}g, x1{s}f, and y1{s}g are pairwise comonotonic, as are the prospects x{t}f1, y{t}g1, x1{t}f1, and y1{t}g1. A preference relation satisfies sign-comonotonic tradeoff consistency if comonotonic tradeoff consistency holds whenever the consequences x, x1, y, y1 are either all non-negative or all non-positive. Theorem: a. Expected utility theory holds iff satisfies TC. b. Cumulative utility theory holds iff satisfies CTC. c. Cumulative prospect theory holds iff satisfies double matching and SCTC. END OF PRESENTATION