What we are going to cover in this article is “log-normal” property of stock prices.

  1. log-normal distribution of stock price: mean and variance of log(S)
  2. Deriving mean and variance of S from its log-normal property

In other words, the article is for someone who has a hard time wrapping their heads around the below formula:

\[ lnS_T \sim \mathrm{N}(lnS_0 + (\mu - \sigma^2/2)T, \sigma^2T) \\ \Rightarrow\\ E(S_T) = S_0e^{\mu T} \\ Var(S_T) = S_0^2e^{2 \mu T}(e^{\sigma^2 T} -1) \]

Related information can be found in fomula 14.4 and 14.5 in 8th edition of Hull’s Derivitatives textbook.
In fact, the authors provide in one of their technical notes how to derive it.
Technical note: Properties of Lognormal Distribution

However, for someone who lack mathematical backgrounds may find it a little hard to parse the meaning.

Before we start, I will assume you already know below:

  1. Geometric Brownian Motion
  1. Log-normal property

\[ lnS_T \sim \mathrm{N}(lnS_0 + (\mu - \sigma^2/2)T, \sigma^2T) \]

  1. basic mathematical statistics skills

Now let’s get started.


  1. There is a randome variable X that follows normal distribtuion. V follows log-normal distribution.
    → that is, log(V) (i.e. X) follows normal distribution. \[ X = ln(V), \ X \sim \mathrm{N}(m, s^2) \]

  2. probability denstiy function(pdf) of normal distribtuion is defined as below:
    \[ f(x): probability \ density \ function \ of \ X \\ f(x) = \frac{1}{\sqrt {2\pi}}*exp[\frac{(x-\mu)^2}{-2s^2}] \]

  3. Variable conversion
    3.1 Definition of variable conversion(one variable case)

\[ v = u(x) \\ x = w(y) \\ (u, v: 1 \ to \ 1 \ corresponding function) \\ \Rightarrow \\ h(v) = f[w(v)]*|J| \\ s.t.\\ f(x): \ pdf \ of \ X \\ h(v): \ pdf \ of \ V \\ J: \ Jacobian \ Matrix \\ where \ J = \frac{d}{dv}w(v) \]

3.2 By applying it to our case, h(v) (i.e. v’s pdf) can be expressed as below:

\[ v = u(x) = e^x \\ x = w(v) = lnv \\ \Rightarrow \ h(v) = f[w(v)]*|J| = \frac{1}{\sqrt{2\pi}\sigma v}exp[\frac{(lnv-\mu)^2}{-2s^2}] \\ (J = |J| = \frac{1}{v}) \]

  1. Deriving moment generating function(mgf) + nth moment
    4.1. First, we need to understand why we derive mgf in the first place. E(V^n), namely nth moment can be computed using mgf. With 1st moment(mean) and 2nd moment(mean of square), Var(V) can also be derived.
    4.2. For a given pdf, h(v), mgf is defined as follows;

\[ <moment \ generating \ function> \\ M_V(t) = E(e^{vt}) \]

4.3. Plus, by definition of mgf, nth moment can be computed as below:

\[ <n^{th} moment>\\ \frac{d^n}{dt^n}M_V(t)|_{t=0} = E(V^n) \]

4.4. That can be rewritten as below:

\[ \int_0^\infty V^nh(v)dv \]

4.5 Since v = e^x, we substitute x for v. Be ware of the change in the range. (x: all real number, v: positive real number)
+ Be careful when subsituting dx for dx.

\[ \int_0^\infty v^n \frac{1}{\sqrt{2\pi}s v} exp[\frac{(lnv-m)^2}{-2s^2}]dv \\ = \int_{-\infty}^\infty exp(nx) \frac{1}{\sqrt{2\pi}s} exp[\frac{(x-m)^2}{-2s^2}]dx \]

4.6 You have to grind away a bit. I leave this part to readers. At the end, you can obtain this:

\[ exp(nm + \frac{n^2s^2}{2}) \int_{-\infty}^\infty \frac{1}{\sqrt{2\pi}s} exp[\frac{(x-m-ns^2)^2}{-2s^2}]dx \]

4.7 But check out the latter part of the outcome. You can see this is just 1 in accordance with the definition of pdf.

\[ \int_{-\infty}^\infty \frac{1}{\sqrt{2\pi}s} exp[\frac{(x-m-ns^2)^2}{-2s^2}]dx \\ \Rightarrow integral \ of \ pdf \ of \ X \sim \mathrm N(m+ns^2, s^2) \ \Rightarrow 1 \]

4.8 Therefore, 4.6 can be rewritten as below. This is V’s nth moment.

\[ \Rightarrow E(V^n) = exp(nm + \frac{n^2s^2}{2}) \]

  1. Mean(1st moement) and Variance(2nd moment - 1st moment^2)
    5.1 1st moment (mean): plug n = 1 in the formula we got from 4.8. \[ E(V) = exp(m + \frac{s^2}{2}) \]

5.2 2nd moment (mean of square): plug n = 2 in the formula we got from 4.9. \[ E(V^2) = exp(2m + 2s^2) \]

5.3 Variance \[ E(V^2) - {E(V)}^2 = Var(V) \]

  1. Now we define X = ln(S), V = S. Then the below comes out. (due to log-normal property of stock price)

\[ lnS_T \sim \mathrm{N}(lnS_0 + (\mu - \sigma^2/2)T, \sigma^2T) \] Then,
\[ m = ln(S_0) + (\mu - \sigma^2/2)T \\ and \\ s = \sigma\sqrt{T} \]

6.1 By pluging m and s we computed in E(V) and V(V) we calculated, we can get the below outcome. (Conclusion)

\[ E(V) = E(S_T) = S_0e^{\mu T} \\ Var(V) = Var(S_T) = S_0^2e^{2\mu T}(e^{\sigma^2 T} -1) \]