Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

The aim of transforming the Black-Scholes PDE is of course to find a form where an relatively easy solution exists. Most of the steps seem to be straightforward - please use this reference: http://planetmath.org/encyclopedia/AnalyticSolutionOfBlackScholesPDE.html

...all but one, actually the last one where a convection-diffusion equation is being transformed into the basic diffusion equation.

[In the article you find it here: "Under the new coordinate system (z), we have the relations amongst vector fields ... leading to the following transformation of equation..."]

The u_x-term vanishes by some magic coordinate transformation. When you look at the derivatives they even seems wrong to me because they state that tau=s and then derive delta/delta tau = delta/delta s + some extra term (delta/delta y * -(r-1/2 sigma^2).

I simply don't get it: first how it works and second how they find that kind of transformation.

share|improve this question
add comment

1 Answer 1

up vote 3 down vote accepted

I'm afraid the Planetmath page put my browser into an infinite reload loop, so I can't help you with the formalism there.

I would recommend instead looking at the change of variables in the Wikipedia article. The last time I checked it, it seemed to work.

Edit: Okay, I have a formulation that works. I'll write s for sigma, so the equation is initially:

Vt + (1/2)s2S2VSS = rV - rSVS.

Since S follows a lognormal random walk (in particular the stochastic diff eq governing S involves a logarithmic derivative), it is natural to change to x = log S, or S = ex, so the log price x follows normal Brownian motion. This yields the equation:

Vt + (1/2)s2(Vxx - Vx) = r(V - Vx).

Black-Scholes is a final-value problem, i.e., we know the value of the option at time T, and diffusion works backwards. It is therefore natural to negate the time variable (and multiply by a suitable scalar to make things neater). tau = (1/2)s2(T-t). Then we get:

(1/2)s2(Vxx - Vx - Vtau) = r(V - Vx).

Finally we rescale the value function to remove exponential growth effects. u = eax + b(tau)V for undetermined coefficients a and b. We can substitute, multiply the equation by 2eax+b(tau)/s2, and we get:

uxx + (something)ux + (something else)u = utau.

(something) is a degree one polynomial in a and is independent of b. (something else) is a degree one polynomial in b, so we can choose a and b to kill those terms. This yields the diffusion equation.

Hope that helps.

share|improve this answer
    
Thank you Scott - anyway it is the same thing here: x=ln(S/K)+(r-sigma^2/2)*tau - I don't understand how you come to this term and how the actual transformation works. p.s.: you are right, there are some browser problems with this site, IE works well though –  vonjd Oct 25 '09 at 6:49
    
I got this solution from Strang's <i>Computational Science and Engineering</i>, which references <i>The Mathematics of Financial Derivatives</i> by Wilmott, Harrison, and Dewynne. –  S. Carnahan Oct 27 '09 at 4:16
    
Scott: Thank you very much, again! Could you please elaborate on this part: "we get: uxx + (something)ux + (something else)u = utau. (something) is a degree one polynomial in a and is independent of b. (something else) is a degree one polynomial in b, so we can choose a and b to kill those terms. " Could you please give an example for "something" and "something else" so that these terms vanish? –  vonjd Oct 27 '09 at 18:09
    
If you work it out, you get: (something) = 2r/s^2 - 1 - 2a, so you can set a = r/s^2 - 1/2 to make it vanish. (something else) = a^2/s^2 - 2ar/s^2 + a - 2r/s^2 + b, so you can set b = 2r/s^2 + 2ar/s^2 - a - a^2/s^2 to make it vanish. –  S. Carnahan Oct 28 '09 at 2:36
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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