Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 38

Compound Utility Theory

(CUT)
(Literature:
(Literature: Paper
Paper Zou
Zou 3)
3)

Two Popular Approaches to


Decision Making under Risk
Expected Utility Theory (EUT)
Von Neumann & Morgenstern (1944)
Many other extensions

Prospect Theory (PT)


Kahneman & Tversky (1979)
PT has grown into cumulative prospect
theory, which blends the initial PT with
rank-dependent models.

2006 by Liang Zou. All rights


reserved.

4-2

Expected Utility Theory (EUT)


EUT says that the objective of decision
making under risk is to maximize ones
expected utility:

EU ( w) p sU ( ws )
s

where w denotes wealth (or


consumption) and ps the probability that
state s will occur.
2006 by Liang Zou. All rights
reserved.

4-3

Prospect Theory (PT)


PT assumes that the utility function (called value
function) depends on gain and loss and has a Sshaped form with a kink at zero (see next slide). Two
implications: loss aversion and risk seeking in the loss
region.
Moreover, PT assumes that people use their own
decision weights (through probability
transformation) instead of true probability p in
calculating the expected value function.
The objective under PT is to maximize

E v

gains 0

s v( gains )

2006 by Liang Zou. All rights


reserved.

losss 0

s v(losss ) u d

4-4

Assumed value function v in


PT
v

loss

2006 by Liang Zou. All rights


reserved.

gain

4-5

Why is CUT useful?


Experimental evidence against EUT. Although EUT
is a normative theory, its scope of application is
limited by peoples systematic violation of its
independence axiom.
Empirical findings that EU models cannot explain:
e.g. equity risk premium puzzle.
PT is empirically more successful. But its
complicated mathematical structure makes it
difficult to apply. Moreover, why do people distort
true probabilities?
Behavioral finance is another way to get things
explained. But its descriptive approach is too
flexible to be useful for rational decision making
2006 by Liang Zou. All rights
under
risk.
reserved.

4-6

Normative, Positive, and


Descriptive Theories
Descriptive theory: purports to explain actual
behavior, allowing irrationality.
Positive theory: purports to explain actual
behavior, assuming people are rational.
Normative theory: purports to advice as to
rational decision making.
It is difficult to agree on a separating line
between these theories. Roughly speaking, the
assumed degree of rationality increases as one
moves toward a normative theory; whereas the
descriptive generality increases as one moves
toward a descriptive theory.
2006 by Liang Zou. All rights
reserved.

4-7

What is rational decision


making?
There is no universal standard for
rationality. Normative theories
typically assume stronger standards of
rationality than descriptive theories
at the cost of loosing some degree of
empirical generality.
Intuitively, rational decision making
suggests optimizing behavior: making
best use of information available and
seeking logical consistency in ones
decision rules.
2006 by Liang Zou. All rights
reserved.

4-8

What kind of theory do we


like?
It depends on what our problems are,
and how we wish to apply the theory.
Most of us prefer theories that help us
make rational decisions as well as
explain economic phenomena.
A normative theory that is incongruent
with reality, or a descriptive theory that
has no predictive power, may be
entertaining but also can be misleading.
2006 by Liang Zou. All rights
reserved.

4-9

Evidence against EUT


The common consequence effect
The common ratio effect
Simultaneous gambling and insurance
These observations have a common name in PT (and
rank-dependent models): attitude toward
probabilities. But this view begs the important
question as to what motivates people to show such an
attitude, especially in situations where the true
probabilities are clearly stated and objectively given.
We shall see that CUT offers an entirely different, yet
appealing, interpretation about behaviors that exhibit
attitude toward risk that the shape of utility functions
cannot explain.
2006 by Liang Zou. All rights
reserved.

4-10

According
Accordingto
to
EUT
you
EUT you
should
shouldhave
have
chosen
choseneither
either
A1
A1and
andB2,
B2,or
or
A2
and
B1.
A2 and B1.
What
Whathave
haveyou
you
chosen?
chosen?

Example (Allais Paradox)

P1: $ 1m

$ 5m (0.10)
P2:

$ 1m (0.89)
$ 0 (0.01)

Which gamble do you prefer? Finding: Most prefer P1.


2006 by Liang Zou. All rights
reserved.

4-11

Allais Paradox (continued)


$ 5m (0.10)

$ 1m (0.11)
P4:

P3:
$ 0 (0.89)

$ 0 (0.90)

Which gamble do you prefer? Finding: Most prefer P4.


2006 by Liang Zou. All rights
reserved.

4-12

Example (ABC Paradox)


1.5
1.2

1.5
0.25

0.5

0.5

1.2

0.25

1
0.5

0.8

1
0.5

0.25
0.25

0.6
(A)

(B)

0.8
0.6

(C)

Assume 1=$10,000. Rank these gambles. Finding: Most for C.


2006 by Liang Zou. All rights
reserved.

4-13

Limitations of extant risk


models

Ad hoc assumptions of risk measures:


Variance
Probability of shortfalls
Semi-variance
Expected loss
Value at risk, etc.

Why choose mean as reward?


Lack of theoretical foundation
2006 by Liang Zou. All rights
reserved.

4-14

The Equity Premium Puzzle


Low Real Risk-free Interest Rate
(1%)
High Real Equity Risk Premium (7%)
Over more than 15 years to date, no
rational theory has been found that
resolves the puzzle.
2006 by Liang Zou. All rights
reserved.

4-15

Compound Utility Theory (CUT)


CUT holds if the objective of decision
under risk is to maximize ones
compound utility under target (or
reference point) :

V (u , d ) V ( EU , ED )

u psU ( ws ), d ps D (ws )
ws

ws

where u and d are called utility reward


and
disutility risk, respectively.
2006 by Liang Zou. All rights
reserved.

4-16

Attractive Features of CUT


It offers a unified approach that is justified by
a set of intuitive axioms.
It gives a general measure of (utility) reward
and (disutility) risk, as well as a simple
framework for analyzing nonlinear
preferences.
It resolves the paradoxes and puzzles in the EU
paradigm in easy and rational terms.
It retains the normative appeal of EUT, and is
easy to be brought to resolve specific decision
problems that call for better understanding of
risk and human attitude toward risk.
2006 by Liang Zou. All rights
reserved.

4-17

Development of CUT
Key Concepts
Axioms
Measure of Disutility Aversion
Eliciting Compound Utility
Resolving Paradoxes

2006 by Liang Zou. All rights


reserved.

4-18

Foundation of CUT
Nature has placed mankind under the governance
of two sovereign masters, pain and pleasure. It is
for them alone to point out what we ought to do, as
well as to determine what we shall do. On the one
hand the standard of right and wrong, on the other
the chain of causes and effects, are fastened to their
throne.
-- Jeremy Bentham (1789)
2006 by Liang Zou. All rights
reserved.

4-19

Key Ideas of CUT


We can agree on a set of rational criteria for
evaluating risk and reward.
Perceptions of risk and reward fundamentally
determine our choice and conduct.
Numerical representation of preference can be
derived via ones evaluation of risk and reward.
People generally avoid disutility-risk. Some
people may demand increasingly higher
marginal compensations of utility-reward for
increases of disutility-risk.
2006 by Liang Zou. All rights
reserved.

4-20

An Example of Theory Making


Suppose we observe the following sequence
1, 2, 5, 10, 17, 26, 37, 50, ...
What theory can we make about the numbers
that will follow?
Step 1: Assume that the observed pattern will
continue.
Step 2: Derive from the assumption in Step 1
a prediction (hypothesis).

2006 by Liang Zou. All rights


reserved.

4-21

According
Accordingto
to
EUT
you
EUT you
should
shouldhave
have
chosen
choseneither
either
A1
A1and
andB2,
B2,or
or
A2
and
B1.
A2 and B1.
What
Whathave
haveyou
you
chosen?
chosen?

General vs. Specific Theories


Consider three theories (hypotheses):
T1: The sequence to follow will be increasing,
starting with a number greater than 50.
T2: Same as T1, and the sequence will have
alternating odd and even numbers.
T3: The sequence is 1+n2, n=8,9,...

T1 is most general, followed by T2,


and T3 is most specific (most easily
wrong). But which theory do you like?
2006 by Liang Zou. All rights
reserved.

4-22

Balancing Generality and


Specificity

Similar to EUT, CUT seeks a specific theory


that has strong predictive power and easy to
apply. Unlike EUT, however, CUT assumes a
two-stage decision: first assessing risk and
reward, then choosing the choice-object with
best risk-reward combination.
Similar to PT, CUT stresses the role of
reference point in ones psychology. But
unlike PT which typically assumes =0, CUT
allows to be a parameter of the decision
maker which may change across individuals
and with the choice environment. More
importantly, CUT does not assume probability
2006 by Liang Zou. All rights
transformation
as PT does.
reserved.

4-23

Choice-objects under risk


In this course, we focus on choiceobjects that are investment
strategies.
Example (price or return
distributions of):
Stocks, bonds, savings,
Options, futures, derivative
contracts
Mutual funds, hedge funds,

2006 by Liang Zou. All rights


reserved.

4-24

Basic Setup
Denote by X the set of outcomes, by the set of
choice-objects that are probability measures on X.

Let f denote an abstract "better than" relation


on , e.g., for any P, Q in, P f Q could mean
"P is preferred to Q", "P is less risky than Q", etc.
Define f , p , p, : as usual.
Partition X and by some X and define X+, X-, +
and - accordingly.
Let denote a probability mixture, i.e., 0.5P 0.5Q
means a lottery that assigns a probability of 0.5 for P
and 0.5 for Q.

2006 by Liang Zou. All rights


reserved.

4-25

Axiom A1: Weak Order


The relation f is complete and transitive on .
That is, for all P, Q, Z in ,
a) Either P ff Q or Q
b) If P ff Q and Q

P (complete)
Z , then P f Z (transitive)

In words, all strategies can be ranked. If


P is better than Q and Q is better than Z,
then P is better than Z.
Note that we have dropped the target
parameter for notational convenience.
2006 by Liang Zou. All rights
reserved.

4-26

Axiom A2: Continuity


For all P, Q, Z in, if P ff Q

Z then

there exist reals 0 < , < 1 such that


P (1 ) Z f Q, P (1 ) Z p Q
In words, if P is better than Q and Q is
better than Z, then a lottery that yields P
with almost sure probability or else Z is
better than Q. Likewise, Q is better than a
lottery that yields Z with almost sure
probability or else P.
2006 by Liang Zou. All rights
reserved.

4-27

Axiom B1: Partial


Independence
For all P, Q, Z in+ (respectively, )
and all real number 0 < < 1, if P f Q
then P (1 ) Z f Q (1 ) Z
In other words, the better than relation is
not affected by the chance to play another
strategy Z. If P is more rewarding (risky) than
Q, then a lottery yielding either P or Z is
more rewarding (risky) than a lottery yielding
either Q or Z (where both lotteries have the
same probabilities for Z ).
2006 by Liang Zou. All rights
reserved.

4-28

Axiom B2: Monotonicity


For all P and Q in , if P ff Q and P
or if P ff Q and P

Q , then P f Q.

In other words, If P is more rewarding


and less risky than Q, then P is better than
Q. If any one of these conditions holds
strictly, then P is strictly better than Q.

2006 by Liang Zou. All rights


reserved.

4-29

Representation Theorem
Axioms A1-A2 and B1-B2 hold true if and
only if there exist functions U, D, and V
such that for all P and Q in ,

P f Q V (u ( P ), d ( P )) V (u (Q ), d (Q ))
u ( P) psU ( ws ), d ( P) ps D ( ws )
ws

ws

where U and D are monotonic, unique up


to a positive ratio scaling, and where V is
continuous, monotonically increasing in u
and decreasing in d, and is unique up to a
positive
transformation.
2006
by Liang Zou.
All rights
reserved.

4-30

V(u,d)

u
d

Example of a compound utility function.


2006 by Liang Zou. All rights
reserved.

4-31

Definition of : A local
measure of disutility aversion
V (u , d )

d
(u, d )
V (u , d )
u
It measures the required marginal increase in
utility-reward that compensates for a marginal
increase in disutility-risk along the indifference
curve.
2006 by Liang Zou. All rights
reserved.

4-32

Graphical: Indifference Curves


u
V(u,d)=constant

d
2006 by Liang Zou. All rights
reserved.

4-33

Comparing CUT with PT


u

Increasing preference

Increasing preference

B
C

B
C

A
d

0
(a) CUT
2006 by Liang Zou. All rights
reserved.

d
(b) PT
4-34

CUT vs. PT: Interpretations


PT: "People appear to have an attitude toward
probabilities, resulting in observed behavior patterns
that only probability transformations seem able to
explain. In the context of decision under risk where true
probabilities are given, the cause for such systematic
distortions of true probabilities is unclear."
CUT: "People generally avoid disutility-risk. Some
people may demand increasingly higher marginal
compensations of utility-reward for increases in
disutility-risk."
Whereas both statements may explain the same
observed (quasiconcave) behavior at a descriptive
level, CUT offers a plausible reason for such effects:
"Because disutility hurts. And for some people the
undesirability of disutility-risk increases more than in
proportion to the desirability of utility-reward."
2006 by Liang Zou. All rights
reserved.

4-35

Resolving Allais Paradox


u

(a)

U(5)

(b)

Q
U(1)

U(5)

P2
P1

Q
P4
P3

P1

Q1
D(0)

2006 by Liang Zou. All rights


reserved.

0 D(1)

P2

P4
P3

Q1
D(0)

d
4-36

A Single Parameter CU
Function
For parsimony, we shall often focus on
the following special CU function:

V (u , d ) U ( )[(1 u ) d ] a 0
a

where U is a usual utility function and


u E max[U ( ws ) / U ( ) 1, 0]
d E max[1 U ( ws ) / U ( ), 0]
2006 by Liang Zou. All rights
reserved.

4-37

Estimating parameter a
In order to derive some idea about the values
of a, consider the ABC paradox. Assume that
the subject evaluates u and d by expected
better and worse outcomes respectively
under target =1. We then find that

A ff C
C ff A

B if 0 < a < 0. 69347


B if 0. 69347 < a < 0. 69958

C ff B
B ff C

A if 0. 69958 < a < 0. 70554


A if 0. 70554 < a

2006 by Liang Zou. All rights


reserved.

4-38

You might also like