A binomial‐tree model provides a discrete‐time representation of how an asset price evolves. At each time step, the price can either move “up” or “down” by a specified factor, capturing a simplified version of market uncertainty. Over multiple steps, this forms a recombining tree of possible paths.
## Discrete Steps
We assume that at each step $t$, the price $S_t$ relates to the price $S_{t-1}$ with a log-return $r_t$.
$ S_t = S_{t-1} *e^{r_t} $
In the binomial tree $r_t$ takes one of two possible values at each step. Either an "up" or "down" return, denoted as follows:
$ r_t = \begin{cases}
\log u, &\text{with probability} \: p \\
\log d, &\text{with probability} \: (1-p)
\end{cases}$
These up and down factors must be calibrated to market data, i.e. the [[Expectation]] and [[Variance]] of the financial asset, which we want to model.
## Bernoulli Formulation
A convenient way to encode these two possible states, is to use a [[Bernoulli Distribution|Bernoulli]] [[Random Variable]] $x_t$.
$ r_t = a+bx_t, \quad r_t = \begin{cases}
\log 1, &\text{with probability} \: p \\
\log 0, &\text{with probability} \: (1-p)
\end{cases}$
In this representation:
- *Upwards case:* When $x_t=1$, which implies that $r_t = a+b$
- *Downwards case:* When $x_t=0$, which implies that $r_t =a$
## Calibrating Model Parameters
The model parameters $a,b$ govern the size of the up and down factors. We want them to reflect the real mean $\mu$ and variance $\sigma^2$ and $p$ of the financial asset over the given time step $t$.
Probability $p$: While there are different ways to estimate the probability of an upwards move, the simplest is to look at historical frequencies.
$ \hat p= \frac{\sum_t\mathbb I_{r_t>0}}{N}$
where:
- Numerator: The count of positive log-returns, expressed as a sum of indicator variables
- Denominator: Number of observed log-returns
Mean:
$
\begin{align}
\mu \equiv \mathbb E[r_t]&=\mathbb E[a+bx_t]\\[2pt]
&= a+b\mathbb E[x_t]\\[2pt]
&= a+bp\\
\hat \mu &=a+b \hat p
\end{align}
$
Variance:
$
\begin{align}
\sigma^2 \equiv \mathrm{Var}(r_t) &= \mathrm{Var}(a+bx_t)\\[2pt]
&= b^2 \mathrm{Var}(x_t)\\[2pt]
&= b^2p(1-p)\\
\hat \sigma^2 &=b^2 \hat p(1-\hat p)
\end{align}
$
Use the established relations, we can solve for $b$:
$
\begin{align}
\sigma^2 &= b^2p(1-p)\\[2pt]
b^2 &= \frac{\sigma^2}{p(1-p)}\\[2pt]
b&=\frac{\sigma}{\sqrt{p(1-p)}}
\end{align}
$
Solving for $a$:
$
\begin{align}
\mu &=a+bp \\[2pt]
a &=\mu-bp \\[2pt]
a &= \mu - \frac{\sigma}{\sqrt{p(1-p)}}*p\\[4pt]
a &= \mu - \sigma\sqrt{\frac{p}{(1-p)}}
\end{align}
$
## Relating to Up/Down Factors
Once $a$ and $b$ are known, we can interpret them in terms of the up & down factors $\log u$ and $\log d$.
Down factor:
$ \log d = a = \mu - \sigma\sqrt{\frac{p}{(1-p)}} $
Up-factor:
$
\begin{align}
\log u &= a+b\\
&= \mu - \sigma\sqrt{\frac{p}{(1-p)}} + \frac{\sigma}{\sqrt{p(1-p)}}\\
&=\mu - \sigma \frac{p}{\sqrt{p(1-p)}}+ \frac{\sigma}{\sqrt{p(1-p)}}\\
&=\mu + \sigma \frac{1-p}{\sqrt{p(1-p)}}
\end{align}
$