From Hull's book when deriving coefficients of up and down movements, $u$ and $d$, of a stock price using binomial tree approach, at some point we get the following equation:

$$e^{\mu\Delta t}(u+d) - ud - e^{2\mu\Delta t} = \sigma^2\Delta t.$$

Then it is stated that from solving the above equation we obtain that $u = e^{\sigma\sqrt{\Delta t}}$ and $d= e^{-\sigma\sqrt{\Delta t}}$. It is also noted that we use Taylor's formula and throwing $\Delta t^2$ and higher terms:

$$e^x = 1 + x + \frac{x^2}{2!} + \frac{x^3}{3!} + \dots.$$

Could you clarify how do we get to this result?

So far I get by using Taylor's formula:

$$e^{\mu\Delta t} \approx 1 + \mu\Delta t,$$ $$e^{2\mu\Delta t} \approx 1 + 2\mu\Delta t.$$

Then the above equation transforms to

$$(1+\mu\Delta t)(u+d) - ud - 1 - 2\mu\Delta t = \sigma^2\Delta t.$$

I am confused how to proceed from here. I tried to do some algebra but it gave no result. For instance, if we assume that $ud=1$ then we get

$$(1+\mu\Delta t)(u+d) - 2(1+\mu\Delta t) = \sigma^2\Delta t,$$ $$(1+\mu\Delta t)(u+d-2) = \sigma^2\Delta t$$ and $$u+d = \frac{\sigma^2\Delta t}{1+\mu\Delta t} + 2$$

Here I am stuck

  • $\begingroup$ note that there are over 30 binomial tree discretizations and this is just one of them, and it is really out of date. $\endgroup$
    – Mark Joshi
    Apr 13 '16 at 3:54

We assume that $u=e^x$ and $d = e^{-x}$. Note that \begin{align*} u &\approx 1+ x +\frac{x^2}{2}, \textrm{ and}\\ d &\approx 1- x +\frac{x^2}{2}. \end{align*} Substituting these into your last equation, \begin{align*} u+d = \frac{\sigma^2 \Delta t}{1+\mu\Delta t} + 2, \end{align*} we obtain that \begin{align*} x^2 \approx \frac{\sigma^2 \Delta t}{1+\mu\Delta t}. \end{align*} Note also that \begin{align*} \frac{1}{\sqrt{1+\mu\Delta t}} \approx 1-\frac{1}{2}\mu\Delta t+\frac{3}{8}(\mu\Delta t)^2. \end{align*} Consequently, \begin{align*} x & \approx \frac{\sigma \sqrt{\Delta t}}{\sqrt{1+\mu\Delta t}}\approx \sigma \sqrt{\Delta t} -\frac{1}{2}\mu\sigma (\Delta t)^{3/2} + \frac{3}{8}\sigma \mu^2 (\Delta t)^{5/2} \approx \sigma \sqrt{\Delta t}. \end{align*} That is, \begin{align*} u &= e^x =e^{\sigma \sqrt{\Delta t}},\\ d &= e^{-x} =e^{-\sigma \sqrt{\Delta t}}. \end{align*}

  • $\begingroup$ Thanks for your answer. Could you explain why $\frac{\sigma\sqrt{\Delta t}}{1 +\frac{1}{2}\mu\Delta t} \approx \sigma\sqrt{\Delta t}$? We assume that $\Delta t$ is small so that denominator is equal to one? Why then we assume $\sqrt{\Delta t}$ not too small so that numerator is not equal to zero? $\endgroup$
    – tosik
    Apr 13 '16 at 6:56
  • 2
    $\begingroup$ Remember that $\sqrt{x}$ tends slower towards zero than $x$ as $x \rightarrow 0 $. $\endgroup$
    – Quantuple
    Apr 13 '16 at 7:48
  • 1
    $\begingroup$ @IvanovNikita, please see my updates. $\endgroup$
    – Gordon
    Apr 13 '16 at 12:57
  • $\begingroup$ How do you obtain u and d without assuming that u = exp(x) and d = exp(-x)? At first glance it is not trivial to understand why we exponeniate only the left side of the equation. Or in other words how do you come to decide to equate u and d to exp (x) and exponeniate only the left side of the equation? $\endgroup$ Aug 11 at 9:21

You can also get this result from Gauss-Hermite quadrature, and this is the subject of a recent paper in the Journal of Derivatives https://jod.pm-research.com/content/early/2021/04/02/jod.2021.1.130.short.

The price of an option can be written as the discounted integral of the payoff $\phi(S)$ times the transition probability density for a lognormal process: $$ C(S,T) = D(T)\int_0^\infty \phi(z)\exp\left(-\frac{\left(\ln\frac{z}{S} - \left(r-\frac{1}{2}\sigma^2\right)T\right)^2}{2\sigma^2T} ​\right)\frac{dz}{z\sqrt{2\pi\sigma^2T}}. $$ Making the substitution $y=\frac{\left(\ln\frac{z}{S} - \left(r-\frac{1}{2}\sigma^2\right)T\right)^2}{\sigma\sqrt{T}}$ leads to an integral amenable to Gauss-Hermite quadrature $$ C(S,T)=D(T)\int_{-\infty}^\infty \phi\left(Se^{\sigma\sqrt{T}y+\left(r-\frac{1}{2}\sigma^2\right)T}\right)e^{-\frac{y^2}{2}}\frac{dy}{\sqrt{2\pi}}. $$ Using second order Gauss-Hermite quadrature leads to $$ C(S,T) \approx \frac{D(T)}{2}\left[\phi\left(Se^{\sigma\sqrt{T} +\left(r-\frac{1}{2}\sigma^2\right)T}\right)+\phi\left(Se^{-\sigma\sqrt{T} +\left(r-\frac{1}{2}\sigma^2\right)T}\right) \right] $$

Now you may think this is very far from the answer, however two things need to be done. First discretize time into timesteps of constant variance by using the Chapman-Komolgorov equation, and second modify Gauss-Hermite quadrature to move the abscissa slightly off the roots of the third Hermite polynomial. The details are in the paper cited above. The result is $$ C(S,T) = D(T)\sum_{j=0}^{N}p^{j}(1-p)^{N-j} {N \choose j}f\left(xe^{(2j-N)\sigma\sqrt{\Delta t}}\right) $$ where $p=\left[\frac{1}{2} + \frac{\left(r-\frac{1}{2}\sigma^2\right)\Delta t}{2\sigma\sqrt{\Delta t}}\right]$. This is equivalent to the probabilities in the original paper by Cox, Ross and Rubinstein (bottom of page 252) Original CRR paper


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.