Limit as $n \to \infty$ for Kelly Criterion

84 Views Asked by At

Suppose that with each game you have probability .7 of winning and .3 of losing. Say you bet half your money in each game. You start with $Y_0$ = 1. Let $Y_n$ be your bankroll after $n$ games and $$ Y_{n+1} = Y_n \cdot W_{n+1}.$$ with $$ W_i = \begin{cases} 1.5 & \text{with probability}~.7 \\ .5 & \text{with probability}~.3 \end{cases}.$$

It follows, $$ Y_n^{1/n} = \left(\prod_{i=1}^n W_i \right)^{1/n} \implies \log Y_n^{1/n} = \frac{1}{n} \sum_{i=1}^n \log W_i = \mathbb{E} \left( \log (W_i) \right) .$$ Now, $$ \mathbb{E} \left(\log W_i \right) = (.7) \log(1.5) + (.3)\log(.5) $$ and hence \begin{align*} ( Y_n)^{1/n} \to e^{ (.7) \log(1.5) + (.3)\log(.5)} &= e^{ (.7) \log(1.5)} \cdot e^{ (.3)\log(.5)} \\ &= (1.5)^{.7} \cdot (.5)^{.3} \\ &= 1.0788. \end{align*}

1

There are 1 best solutions below

1
On BEST ANSWER

This is not a valid application of the strong law of large numbers... what's the sample mean here? Also what is $W_i^n$ ... there are $n$ different possible values of $i$ here... you can only take them to be all the same after you take the expectation value.

The trick here is to take logs. You have $Y_n = \prod_{i=1}^n W_i$ so $$ \log Y_n^{1/n} = \frac{1}{n} \sum_{i=1}^n \log(W_i).$$ Now, this is a sample average of $\log(W),$ so is something the law of large numbers applies to since $\log(W_i)$ are i.i.d. and integrable. The mean of $\log(W)$ is $$ \frac{1}{2}\log(1.5) + \frac{1}{2}\log(0.5) = \frac{1}{2}\log(3/4),$$ so $$ (Y_n)^{1/n}\to e^{\frac{1}{2}\log(3/4)} = \sqrt{3/4}. $$

Note that this means, somewhat counterintuitively, that $Y_n\to 0$ almost surely, even though $E(Y_n)\to 1.$ In fact, by inreasing the win probability slightly, you can find an example where $E(Y_n)\to \infty$ but $Y_n\to 0$ almost surely.

Edit whoops, I didn't see that the probabilities of going up vs down were $.7$ and $.3,$ not 50-50. I will leave it up to you to plug those in. Since this is purportedly the Kelly criterion, I'm guessing $Y_n$ will converge to $\infty,$ not zero, almost surely, i.e. the answer will be $>1.$

Edit 2 Actually, the Kelly criterion for these odds is betting $40\%,$ but betting $50\%$ isn't too greedy and still gives positive growth.