Value of experimentation on decision analysis problem?

223 Views Asked by At

The following table represents a decision analysis problem (in units of a thousand dollars)

enter image description here

Suppose you can obtain information which will tell you, with certainty, whether or not state 1 (S1) will occur. What is the maximum amount you would be ready to pay to obtain the information about S1? (Using Bayes' Rule)

I already calculated the expected payoffs of each Alternative w/o the perfect information which is E[A1] = $33,000 E[A2] = $29,000 E[A3] = $39,000.

Not really sure how to calculate the expected total payoff with or without the S1 information, or how to determine it's worth? Thank you.

1

There are 1 best solutions below

0
On

You have access to information that can discriminate for sure whether nature $S$ is in state $s_1$ or not. You can think of this as a conditional probability mass function $q$ over a piece of data $D$ that assumes either value 0 (if $S\neq s_1$) or 1 (if $S=s_1$), e.g. $$ q(D=1 \mid S=s) = \begin{cases} 1 & \text{if } s=s_1 \\ 0 & \text{if } s\neq s_1 \end{cases}. $$ and similarly for $q(D=0 \mid S=s)$.

You may now apply Bayes' rule to determine what is your updated information regarding the state of nature given either $D=0$ or $D=1$. Both of these will have some expected value $U_0$ or $U_1$ computed similarly to the prior information.

Note that both $D=0$ and $D=1$ have a prior probability $P(D=d)$, so the expected utility before actually obtaining the information is $$ P(D=0) U_0 + P(D=1) U_1. $$

The price you should be willing to pay for this information should not exceed the improvement in the expected value above compared to the expected value of the prior information.