9.3: St. Petersburg two envelope problem
This section is extracted from
the following:
$(\sharp):$ | S. Ishikawa; The two envelopes paradox in non-Bayesian and Bayesian statistics $\quad$ ( arXiv:1408.4916v4 [stat.OT] 2014 ) |
Now, we will review the St. Petersburg two envelope problem (cf. D.J. Chalmers, "The St. Petersburg Two-Envelope Paradox," Analysis, Vol.62, 155-157, (2002)).
$(\sharp):$ | a coin was flipped until it came up heads, and if it came up heads on the $k$-th trial, $2^k$ is put into the envelope. This procedure is performed separately for each envelope. |

[(P2):Why is it paradoxical?].
You reason that, before opening the envelopes A and B, the expected values $E(x)$ and $E(y)$ in A and B is infinite respectively. That is because \begin{align} 1 \times \frac{1}{2}+ 2 \times \frac{1}{2^2}+ 2^2 \times \frac{1}{2^3}+ \cdots =\infty \end{align}

(P2): St. Petersburg two envelope problem: classical mixed measurement
Define the state space $\Omega$ such that $\Omega=\{\omega= 2^k \; |\; k=1,2, \cdots \}$, with the discrete metric and the counting measure $\nu$. And define the exact observable ${\mathsf O}=(X, {\mathcal F}, F )$ in $L^\infty(\Omega, \nu )$ such that
\begin{align*} & X= \Omega, \quad {\mathcal F}=2^X \equiv \{ \Xi \;|\; \Xi \subseteq X \} \\ & [F(\Xi)](\omega) = \chi_{{}_\Xi} (\omega ) \equiv \begin{cases} 1 \quad & ( \omega \in \Xi ) \\ 0 \quad & (\omega \notin \Xi ) \end{cases} \qquad (\forall \Xi \in {\mathcal F}, \forall \omega \in \Omega ) \end{align*}Define the mixed state $w$ ($\in L^1_{+1}(\Omega, \nu )$, i.e., the probability density function on $\Omega$) such that
\begin{align*} w_0(\omega) = 2^{-k} \quad (\forall \omega = {2^k} \in \Omega ). \end{align*}Consider the mixed measurement ${\mathsf M}_{L^\infty (\Omega, \nu )} ({\mathsf O}=(X, {\mathcal F}, F ), {\overline S}_{[\ast]}(w_0) )$. Axiom${}^{{\rm (m)}}$ 1(C$_1$) says that
$(A):$ | the probability that a measured value $ 2^k $ is obtained by ${\mathsf M}_{L^\infty (\Omega )} ({\mathsf O}=(X, {\mathcal F}, F ),$ $ {\overline S}_{[\ast]}(w_0))$ is given by $ 2^{-k} $. |
Note that you knew that A contained $x=2^m$ dollars (and thus, $E =\infty > 2^m $). There is a reason to consider that the switching to $B$ is an advantage.
Remark 9.9 After you get a measured value $2^m$ from the envelope $A$, you can guess (also see Bayes theorem later) that the probability density function $w_0$ changes to the new $w_1$ such that $ w_1 ( 2^m) = 1, w_1 ( 2^k) = 0 ( k \not= m )$. Thus, now your information about $A:w_1$ and $B:w_0$ is not symmetrical (under Bayesian statistics (cf. $\S$9.4)). Hence, in this case, it is true: "The Other Person's envelope is Always Greener".
$\fbox{Note 9.2}$ |
There are various criterions except the expectaion.
For example, consider the criterion such that
|