Polya’s Urn and conditional probability

58 Views Asked by At

Setting:

At time $0$, an urn contains $1$ black ball and $1$ white ball. At each time $n ≥ 1$ a ball is chosen uniformly at random from the urn and is replaced together with a new ball of the same colour. Just after time $n$, there are therefore $n + 2$ balls in the urn, of which $B_n + 1$ are black, where $B_n$ is the number of black balls chosen by time $n$. We let $F_n = σ(B_1,...,B_n)$ for $n ≥ 1$ and $F_0 = \{∅,Ω\}$.

I want to show: $\mathbb{P}(B_{n+1}=b | \mathcal{F}_n)=\frac{B_n +1}{n+2}\chi_{B_n = b-1}+ \frac{n+1-B_n}{n+2}\chi_{B_n = b} $

I think that this comes from: $\mathbb{P}(B_{n+1}=b | \mathcal{F}_n)=\mathbb{P}(B_{n+1}=b | B_{n}=b-1)\chi_{B_n = b-1}+ \mathbb{P}(B_{n+1}=b | B_{n}=b)\chi_{B_n = b} $ This of course intuitively makes sense. But why is it true formally?

I tried to state a general version of this:

Let $X$ be a real, discrete random variable taking values $\{x_1, \dots, x_n\}$. Let $\mathcal{F}$ be a $\sigma$-algebra. If there is a partition $\{F_i\}_i \subset \mathcal{F}$ of $\Omega$, we have: \begin{equation} \mathbb{E}(\chi_{X=x_i} | \mathcal{F})=\sum_{j=1}^n \mathbb{E}(\chi_{X=x_i} | F_j) \chi_{F_j} \end{equation}

I tried to show this via the unique property $\forall B \in \mathcal{F}: | \mathbb{E}(\chi_B \chi_{X=x_i})= \mathbb{E}(\chi_B \sum_{j=1}^n \mathbb{E}(\chi_{X=x_i} | F_j) \chi_{F_j}) $. But I failed, can someone maybe tell me if my general statement is true or where the property above comes from. Thank you!

1

There are 1 best solutions below

2
On BEST ANSWER

The general statement is false, which can be easily seen if you take $\{\Omega\}$ as the partition. The right side then reduces to $\mathbb E(\chi_{X=x_i}|\Omega)\chi_{\Omega} = \mathbb E[\chi_{X=x_i}]$ and it's obviously not true that every random variable is equal to its expectation.

In fact, you can show that your right side is just $\mathbb E(\chi_{X=x_i}|F_1,\dots, F_n)$, conditioning on the $\sigma$ algebra generated by the partition, so the statement you want to show and "intuitively makes sense" in your question is actually $$ \mathbb P(B_{n+1}=b|\mathcal F_n) = \mathbb P(B_{n+1}=b|B_n). $$ You want to show that $B_{n+1}$ only depends on $B_n$, and not on the rest of $\mathcal F_n$. This property actually has a name: It is called the Markov property and what fulfils this property is called a Markov chain. But actually it is easier to proof the original statement. Just note that according to the text, the ball chosen in step $n+1$ only depends on the number of balls in the urn. This means $$ \mathbb P(B_{n+1}-B_{n} = 1 |\mathcal F_n) = \frac{B_n+1}{n+2} $$ a.s. and we can get the statement as \begin{align} \mathbb P(B_{n+1} = b |\mathcal F_n) &= \mathbb P(B_n=b, B_{n+1}-B_{n} = 0 |\mathcal F_n) +\mathbb P(B_n=b-1, B_{n+1}-B_{n} = 1 |\mathcal F_n) \\ &= \chi_{B_n=b}\mathbb P(B_{n+1}-B_{n} = 0 |\mathcal F_n) +\chi_{B_n=b-1}\mathbb P(B_{n+1}-B_{n} = 1 |\mathcal F_n) \\&= \frac{B_n+1}{n+2}\chi_{B_n=b} + \frac{n+1-B_n}{n+2}\chi_{B_n=b-1}. \end{align} In the second step, we used the fact that for some $\sigma$ algebra $\mathcal F$, a $\mathcal F$-measurable variable $Y$ can be factored out as $\mathbb E(XY|\mathcal F) = Y\mathbb E(X|\mathcal F)$ a.s.

Edit: If we now turn back to the Markov property, we see that the right side of the equation only depends on $B_n$, so $\mathbb E(B_{n+1}| \mathcal F_n)$ is $\sigma(B_n)$-measurable and thus equal to $\mathbb E(B_{n+1}| B_n)$. If you can't see why $\sigma(B_n)$-measurability implies that, see that \begin{align} \mathbb E(B_{n+1}=b|B_n) &= \mathbb E\left[\mathbb E(B_{n+1}=b|\mathcal F_n)|B_n\right] \\ &= \mathbb E\left[\frac{B_n+1}{n+2}\chi_{B_n=b} + \frac{n+1-B_n}{n+2}\chi_{B_n=b-1} | B_n \right] \\ &= \frac{B_n+1}{n+2}\chi_{B_n=b} + \frac{n+1-B_n}{n+2}\chi_{B_n=b-1} \\ &= \mathbb E(B_{n+1}=b|B_n). \end{align}

Edit 2: That the ball drawn in step $n$ does only depend on $B_n$ in $\mathcal F_n$ is only written in the text or rather only implied. We assume that a uniformly random ball is chosen, that means, all balls have the same probability to be picked, and that this is not dependent on the balls picked before ($\mathcal F_n$).