DECISION THEORY

Decision theory is a body of knowledge and related analytical techniques of different degrees of
formality designed to help a decision maker choose among a set of alternatives in light of their
possible consequences. Decision theory can apply to conditions of certainty, risk, or uncertainty.
[decision UNDER certainty] means that each alternative leads to one and only one consequence,
and a choice among alternatives is equivalent to a choice among consequences. In [DECISION
UNDER risk] each alternative will have one of several possible consequences, and the
probability of occurrence for each consequence is known. Therefore, each alternative is
associated with a probability distribution, and a choice among probability distributions. When
the probability distributions are unknown, one speaks about [DECISION UNDER uncertainty.]
Decision theory recognizes that the ranking produced by using a criterion has to be consistent
with the decision maker's objectives and preferences. The theory offers a rich collection of
techniques and procedures to reveal preferences and to introduce them into models of decision. It
is not concerned with defining objectives, designing the alternatives or assessing the
consequences; it usually considers them as given from outside, or previously determined. Given
a set of alternatives, a set of consequences, and a correspondence between those sets, decision
theory offers conceptually simple procedures for choice. In a decision situation under certainty
the decision maker's preferences are simulated by a single-attribute or MULTIATTRIBUTE
VALUE FUNCTION that introduces ordering on the set of consequences and thus also ranks the
alternatives. Decision theory for risk conditions is based on the concept of utility (see utility,
sense 2). The decision maker's preferences for the mutually exclusive consequences of an
alternative are described by a utility function that permits calculation of the EXPECTED
UTILITY for each alternative. The alternative with the highest expected utility is considered the
most preferable. For the case of uncertainty, decision theory offers two main approaches. The
first exploits criteria of choice developed in a broader context by game theory, as for example the
[MAX-MIN RULE,] where we choose the alternative such that the worst possible consequence
of the chosen alternative is better than (or equal to) the best possible consequence of any other
alternative. The second approach is to reduce the uncertainty case to the case of risk by using
SUBJECTIVE PROBABILITIES, based on expert assessments or on analysis of previous
decisions made in similar circumstances. See also: game theory, optimization, utility, value
(IIASA)

What
is
Gam
e
Theo

"You treat world history as a mathematician does mathematics, in which nothing but laws and formulae exist, no reality,
no good and evil, no time, no yesterday, no tomorrow, nothing but an eternal shallow, mathematical present." Otto Hess

ry?
by David K.
Levine,
Department of
Economics,
An Instructive
Example
Readings About
Game Theory

What economists call game theory psychologists call the theory of social
situations, which is an accurate description of what game theory is about.
Although game theory is relevant to parlor games such as poker or bridge,
most research in game theory focuses on how groups of people interact.
There are two main branches of game theory: cooperative and
noncooperative game theory. Noncooperative game theory deals largely
with how intelligent individuals interact with one another in an effort to
achieve their own goals. That is the branch of game theory I will discuss
here.
In addition to game theory, economic theory has three other main branches:
decision theory, general equilibrium theory and mechanism design theory.
All are closely connected to game theory.
Decision theory can be viewed as a theory of one person games, or a game
of a single player against nature. The focus is on preferences and the
formation of beliefs. The most widely used form of decision theory argues
that preferences among risky alternatives can be described by the
maximization the expected value of a numerical utility function, where
utility may depend on a number of things, but in situations of interest to
economists often depends on money income. Probability theory is heavily
used in order to represent the uncertainty of outcomes, and Bayes Law is
frequently used to model the way in which new information is used to
revise beliefs. Decision theory is often used in the form of decision
analysis, which shows how best to acquire information before making a
decision.
General equilibrium theory can be viewed as a specialized branch of game
theory that deals with trade and production, and typically with a relatively
large number of individual consumers and producers. It is widely used in
the macroeconomic analysis of broad based economic policies such as
monetary or tax policy, in finance to analyze stock markets, to study
interest and exchange rates and other prices. In recent years, political
economy has emerged as a combination of general equilibrium theory and
game theory in which the private sector of the economy is modeled by
general equilibrium theory, while voting behavior and the incentive of
governments is analyzed using game theory. Issues studied include tax
policy, trade policy, and the role of international trade agreements such as

the European Union.

Mechanism design theory differs from game theory in that game theory
takes the rules of the game as given, while mechanism design theory asks
about the consequences of different types of rules. Naturally this relies
heavily on game theory. Questions addressed by mechanism design theory
include the design of compensation and wage agreements that effectively
spread risk while maintaining incentives, and the design of auctions to
maximize revenue, or achieve other goals.

An Instructive Example
One way to describe a game is by listing the players (or individuals)
participating in the game, and for each player, listing the alternative choices
(called actions or strategies) available to that player. In the case of a twoplayer game, the actions of the first player form the rows, and the actions of
the second player the columns, of a matrix. The entries in the matrix are
two numbers representing the utility or payoff to the first and second player
respectively. A very famous game is the Prisoner's Dilemma game. In this
game the two players are partners in a crime who have been captured by the
police. Each suspect is placed in a separate cell, and offered the opportunity
to confess to the crime. The game can be represented by the following
matrix of payoffs
not
confess

confess

not
confess

5,5

0,10

confess

10,0

1,1

Note that higher numbers are better (more utility). If neither suspect
confesses, they go free, and split the proceeds of their crime which we
represent by 5 units of utility for each suspect. However, if one prisoner
confesses and the other does not, the prisoner who confesses testifies
against the other in exchange for going free and gets the entire 10 units of
utility, while the prisoner who did not confess goes to prison and gets
nothing. If both prisoners confess, then both are given a reduced term, but
both are convicted, which we represent by giving each 1 unit of utility:
better than having the other prisoner confess, but not so good as going free.
This game has fascinated game theorists for a variety of reasons. First, it is
a simple representation of a variety of important situations. For example,
instead of confess/not confess we could label the strategies "contribute to
the common good" or "behave selfishly." This captures a variety of

situations economists describe as public goods problems. An example is the

construction of a bridge. It is best for everyone if the bridge is built, but
best for each individual if someone else builds the bridge. This is
sometimes refered to in economics as an externality. Similarly this game
could describe the alternative of two firms competing in the same market,
and instead of confess/not confess we could label the strategies "set a high
price" and "set a low price." Naturally is is best for both firms if they both
set high prices, but best for each individual firm to set a low price while the
opposition sets a high price.
A second feature of this game, is that it is self-evident how an intelligent
individual should behave. No matter what a suspect believes his partner is
going to do, is is always best to confess. If the partner in the other cell is
not confessing, it is possible to get 10 instead of 5. If the partner in the
other cell is confessing, it is possible to get 1 instead of 0. Yet the pursuit of
individually sensible behavior results in each player getting only 1 unit of
utility, much less than the 5 units each that they would get if neither
confessed. This conflict between the pursuit of individual goals and the
common good is at the heart of many game theoretic problems.
A third feature of this game is that it changes in a very significant way if the
game is repeated, or if the players will interact with each other again in the
future. Suppose for example that after this game is over, and the suspects
either are freed or are released from jail they will commit another crime and
the game will be played again. In this case in the first period the suspects
may reason that they should not confess because if they do not their partner
will not confess in the second game. Strictly speaking, this conclusion is
not valid, since in the second game both suspects will confess no matter
what happened in the first game. However, repetition opens up the
possibility of being rewarded or punished in the future for current behavior,
and game theorists have provided a number of theories to explain the
obvious intuition that if the game is repeated often enough, the suspects
ought to cooperate.
If you wish to learn more about game theory, there a variety of good books
on the topic.

OPTIMIZATION
Optimization is an activity that aims at finding the best (i.e., optimal) solution to a problem. For
optimization to be meaningful there must be an OBJECTIVE FUNCTION (see below) to be
optimized and there must exist more than one FEASIBLE SOLUTION, i.e., a solution which
does not violate the constraints. The term optimization does not apply, usually , when the number

of solutions permits the best to be chosen by inspection, using an appropriate criterion (see
DECISION THEORY). One distinguishes SINGLE OBJECTIVE and MULTIOBJECTIVE
optimization. In the first case, the objective is SCALAR-VALUED (it can be measured by a
single number); in the second, the objective is VECTOR-VALUED (its value is expressed by an
n-tuple of numbers). In mathematical terms, the formulation of an optimization problem involves
decision variables, Xl, X2,...Xn, the objective function,
Q = f(Xl,X2,..Xn)
constraint relations, usually of the form
Gi(Xl,X2,...Xn) greater than or equal to O, i = l,2,...m.
The OPTIMAL SOLUTION (or "solution to the optimization problem") is values of decision
variables xl, x2,...xn that satisfy the constraints and for which the objective function attains a
maximum (or a minimum, in a minimization problem). Very few optimization problems can be
solved analytically, that is, by means of explicit formulae. In most practical cases appropriate
computational techniques of optimization (numerical procedures of optimization) must be used.
Among those techniques LINEAR PROGRAMMING permits the solution of problems in which
the objective function and all constraint relations are linear. NONLINEAR PROGRAMMING
does not have this restriction, but can manage many fewer decision variables and constraints.
INTEGER PROGRAMMING serves to solve problems where the decision variables can take
only integer values. stochastic or PROBABILISTIC PROGRAMMING must be used for
problems where the objective function or constraint relations contain random-valued parameters
(in the latter case, the problem is referred to as a CHANGE-CONSTRAINED PROBLEM). A
special case is dynamic optimization problems where the decision variables are not real numbers
or integers but functions of one or more independent variables -- functions of time or space
coordinates, for example. Dynamic optimization problems are sometimes referred to as "optimal
control problems." There exist special techniques to solve such problems; they often make use of
DISCRETIZATION of the independent variables, for example dividing the time axis into a
number of intervals and considering the solutions to be constant over those intervals. A singleobjective optimization problem may have (and usually does have) a single-valued, unique
solution. The solution to a multiobjective problem is, as a rule, not a particular value, but a set of
values of decision variables such that, for each element in this set, none of the objective
functions can be further increased without a decrease of some of the remaining object functions
(every such value of a decision variable is referred to as pareto-optimal). (IIASA)

VALUE
Value can be either objective or subjective. In the latter case, it means subjective worth or
importance. For example, "the value of future benefits to thedecision maker," "the value of clean
air to the society." For the purposes of analysis, the subjective values must be measured on some
scale. These measures of value should be based on preferences expressed by the person or group
of interest.

In [VALUE ANALYSIS,] one considers that the value v is related to the physical or other
objective measure y of a consequence by a subjectively defined [VALUE FUNCTION,] so that v
= f(y). A value function usually departs from proportionality, i.e.,it usually is a nonlinear
dependence. A typical example is the subjective value of money to an individual: the first l,000
schillings in his savings account are probably of more value to him that the l,000 schillings that
would increase the state oh his account from 100,000 to 101,000 schillings. The value of a
multiattribute consequence with VALUE-RELEVANT ATTRIBUTES y1,y2,..yn can be
expressed by a MULTIATTRIBUTE VALUE function, v(yl,y2,..yn). A multiattribute value
function must satisfy the following condition: v(yl,y2,..yn) is greater than or equal to
v(y'l,y'2,..y'n)
if and only if the multiattribute consequence (yl,y2,..yn) is preferred or indifferent to
(y'1,y'2..,y'n).
Several theories exist according to which a multiattribute value function V(.) can, in appropriate
cases, be expressed as an aggregate of single-attribute functions Vi(.). For example, the additive
[CONJOINT measurement theory] assumes that
n
v(yl,y2,..,yn) = SUM Vi(yi).
i=l