Suppose you have a stock, its initial price is $1$. After a period of time $\Delta t$, its price will either get multiplied by $1+\sqrt{\Delta t}$ (goes up), or get multiplied by $1-\sqrt{\Delta t}$ (goes down). Both case has $1/2$ probability. Now let $n\Delta t=1$, What is the probability of the stock price being $x$ at $t=1$ for large $n$?

Here is my idea:

Well, the probability $P(x)$ is only non-zero for certain $x$. The stock price can go up $a$ times with $0 \leq a \leq n$, and $a$ being an integer. If the stock price goes up $a$ times, then its price would be $(1+\frac{1}{\sqrt{n}})^a(1-\frac{1}{\sqrt{n}})^{n-a}$, so the set of possilbe prices would be $X=\{ (1+\frac{1}{\sqrt{n}})^a(1-\frac{1}{\sqrt{n}})^{n-a}\}^n_{a=0}$, each with probability ${\frac{n \choose a}{2^n}}$.

When $n$ gets larger and later, $X$ will get denser and denser on $\Bbb R$, eventually every real number will be a possible price up to a small error $\epsilon$ (can be made arbitrarily small by choosing some $n$).

Observe the identity $\lim_{n\to\infty} (1+\frac{1}{\sqrt{n}})^{\frac{n}{2}+\sqrt{n}r} (1-\frac{1}{\sqrt{n}})^{\frac{n}{2}-\sqrt{n}r}=e^{2r-\frac{1}{2}}$, this means to achieve a stock price $e^{2r-\frac{1}{2}}$, the stock needs to go up ${\frac{n}{2}+\sqrt{n}r}$ times. In other words, let $x=e^{2r-\frac{1}{2}}$, the stock price will be $x$ if it goes up $a={\frac{n}{2}+\frac{\sqrt{n}}{2}(\ln x+\frac{1}{2})}$ times (choose appropriate $n$ and $x$ so that $a$ is an integer).

Consider $P(X|x\leq X \leq x+\Delta x)$, for small $\Delta x$. It should be $P(X=x)$ times $M$=number of possible prices between $x$ and $x+\Delta x$. Where $M=\Delta a=\frac{\sqrt{n}}{2x}\Delta x$

Putting everything together, $P(X|x\leq X \leq x+\Delta x)$ should be $\lim_{n\to\infty}\frac{\sqrt{n} {n \choose {\frac{n}{2}+\frac{\sqrt{n}}{2}(\ln x+\frac{1}{2})}}\Delta x}{2x2^n}=\frac{e^{-\frac{(\ln x-\frac{1}{2})^2}{2}}\Delta x}{x^2\sqrt{2\pi}}$ enter image description here

Which means the PDF is:

$\rho(x)=\frac{e^{-\frac{(\ln x-\frac{1}{2})^2}{2}}}{x^2\sqrt{2\pi}},x>0$.

This is just a log-normal distribution.


You can also apply the CLT.

The sum of a large number of random variables tends towards a normal distribution. Here you are taking the product of a large number of random variables. Take log on both sides and apply the central limit theorem.

Then $Y\sim log(X)$ has a normal distribution, and therefore $X\sim \text{lognormal}$

PS: Your working looks correct to me.

  • $\begingroup$ I see. X~lognormal(-1/2,1), the -1/2 is actually coming from the second term of the taylor expansion of ln(1+x). $\endgroup$ – Mango Apr 17 at 17:47
  • $\begingroup$ sorry quick question. On going over your working again, I found $\lim_{n\to\infty}\frac{\sqrt{n} {n \choose {\frac{n}{2}+\frac{\sqrt{n}}{2}(\ln x+\frac{1}{2})}}\Delta x}{2x2^n}=\frac{e^{-\frac{(\ln x-\frac{1}{2})^2}{2}}\Delta x}{x^2\sqrt{2\pi}}$ Why does the $x$ on the LHS simplify to $x^2$ in the denominator on the RHS? Note the pdf of the lognormal: en.wikipedia.org/wiki/Log-normal_distribution corresponds to $x$ in the denominator $\endgroup$ – Rahul Madhavan Apr 17 at 18:01
  • $\begingroup$ as regards your question in the comments, we are summing over, $\frac{1}{2}ln(1+x)+\frac{1}{2}ln(1-x)$, which cancel out in the first term, and as you point out the second terms add up (we ignore the 4th and higher order terms), so yes $\endgroup$ – Rahul Madhavan Apr 17 at 18:12
  • $\begingroup$ I think RHS simplifies to lognormal(-1/2,1) $\endgroup$ – Mango Apr 17 at 19:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.