Scribd
Upload
117 views2 pages

Subjective Probability

This document discusses subjective probability and how it relates to rational decision making under uncertainty. It presents a model where: 1) A decision maker has preferences over gambles with uncertain outcomes that satisfy basic rationality properties. 2) The decision maker behaves as if they assign subjective probabilities to each possible outcome of an experiment. 3) These subjective probabilities are uniquely defined in a way that makes the decision maker's preferences consistent with maximizing expected utility, where utility is defined over the possible payoffs and subjective probabilities are used to calculate expected values.

Uploaded by

Tim
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
117 views2 pages

Subjective Probability

This document discusses subjective probability and how it relates to rational decision making under uncertainty. It presents a model where: 1) A decision maker has preferences over gambles with uncertain outcomes that satisfy basic rationality properties. 2) The decision maker behaves as if they assign subjective probabilities to each possible outcome of an experiment. 3) These subjective probabilities are uniquely defined in a way that makes the decision maker's preferences consistent with maximizing expected utility, where utility is defined over the possible payoffs and subjective probabilities are used to calculate expected values.

Uploaded by

Tim
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Appendix 1, Part 2. Subjective Probability.

We have assumed that the rational decision maker knows the exact probabilities of any lottery, p P , that is oered to him, as he would for example if the lottery were conducted using a roulette wheel. The more general situation also arises in which the probability distribution can only be guessed at, as for example when the payos are awarded according to the outcome of a horse race. We do not refer to such situations as a lottery, but rather as a gamble. We assume that our rational person will also have a preference on gambles that satises A1 and A2. The question arises: When will the person act as if he had assigned probabilities to the possible outcomes of the horse race and was trying to maximize his expected utility? Such an assignment of probabilities may be considered as that persons subjective probability or personal probability of the outcomes of the race. This approach to subjective probability is due to Anscombe and Aumann (1963) Ann. Math. Statist. 34, 199-205. The situation is modelled as follows. We start with a set of payos, P, the set P of nite probability distributions on P, and a preference pattern, , on P that satises A1 and A2. Let u denote a utility on P that agrees with . Now consider an experiment with nite number, m, of outcomes 1 , 2 , . . . , m , one and only one of which is bound to occur. For arbitrary p1 , p2 , . . . , pm in P , we use [p1 , p2 , . . . , pm ] to denote the gamble whose payo is pj if j occurs, j = 1, . . . , m. Let G denote the set of all such gambles and let G denote the set of all nite probability distributions on G. An element, g, of G that gives weight to just k elements of G has the form, g = (1 [p11 , . . . , p1m ], . . . , k [pk1 , . . . , pkm ]) where the [pi1 , . . . , pim ] are elements of G, and 1 , . . . , k are probabilities adding to one. We assume that there is a preference relation, g , on G that satises A1 and A2. Then by Theorem 1 there exists a utility ug on G that satises the analog of (2), namely g1 g2 if and only if ug (g1 ) ug (g2 ), and
g.

(6) Let pi ,

We make three assumptions relating the two preference relations, pi , pij , etc., denote arbitrary elements of P. A3. If pi pi , then [p1 , . . . , pi1 , pi , pi+1 , . . . , pm ]
g

[p1 , . . . , pi1 , pi , pi+1 , . . . , pm ].

A4. If p p , then [p, . . . , p] g [p , . . . , p ]. A5. (1 [p11 , . . . , p1m ], . . . , k [pk1 , . . . , pkm ])


g

[(1 p11 , . . . , k pk1 ), . . . , (1 p1m , . . . , k pkm )].

Assumptions A3 and A4 say that the preference relation on P carries over onto the preference relation g on G. Assumption A5 says that every element of G is equivalentg to the element of G in which the randomization provided by the j is performed after the outcome of the experiment is observed. As Anscombe and Aumann express it, if the payo is to be determined by a roulette wheel and a horse race, the decision maker is indierent whether the roulette wheel is spun before or after the race. A1

Theorem 2. If both preference relations on P and g on G satisfy A1 and A2, and if Assumptions A3, A4, and A5 are satised, then there exist utilities u on P satisfying (2) and ug on G satisfying (6), such that there exist probabilities 1 , . . . , m adding to one with (7) ug [p1 , . . . , pm ] = 1 u(p1 ) + + m u(pm ). Furthermore, the i are uniquely determined provided there exist p and p in P such that pp. This theorem may be interpreted as follows. If a decision maker has preferences satisfying the conditions of the theorem, then he behaves as if it were known that the probability of outcome i is i , i = 1, . . . , m, and his preferences among gambles agrees with preferring gambles with higher expected utility. Thus, i can be considered as his subjective probability, or personal probability, of outcome i . For a proof of Theorem 2, see the paper of Anscombe and Aumann, or Section 1.4 of Ferguson (1968). It might be helpful to see how the i are constructed. Let q and q be any elements of P such that q q . Then from A4, [q, . . . , q] g [q , . . . , q ]. Since u and ug are determined only up to change of location and scale, we may choose u so that u(q) = 0 and u(q ) = 1, and ug so that ug [q, . . . , q] = 0 and ug [q , . . . , q ] = 1. Then i may be dened as i = ug [q, . . . , q , . . . , q] with q in the ith coordinate and qs elsewhere. One then checks that the i do not depend on the choice of q and q , and that (7) is satised in general.

A2

You might also like