I wanted to figure how how much faster the Sobol quasi random numbers convergence to the B&S call price compared with pseudo random numbers. To generate the Sobol numbers I used the randtoolbox in R to generate these numbers. When using just one step, so from t=0 to t=T, it’s easy. I used the following formula to go from s(0) to s(T).

S_t= S_0*exp⁡((μ- σ^2/2)*t+ σW_t, where W_t is a Sobol Random Number

I use the Sobol numbers and therefore the convergence is much faster because these numbers are better normal distributed when using pseudo random numbers.

My problem is the following:

How do I need to generate these numbers if I use intermediate steps in my simulation, do I need to use more dimensions or just generate more Sobol numbers from the same numbers. I’m already stuck for a long time on this. Hopefully somebody can help me, especially using the randtoolbox package from R to generate these numbers.


  • $\begingroup$ I have a very detailed chapter on this in More Mathematical Finance as Mathias Korner already pointed out. You have to be really careful when you use Sobol. Treating them like pseudo-randoms is a recipe for disaster. $\endgroup$
    – Mark Joshi
    Nov 5, 2015 at 23:25

2 Answers 2


First let me say that in the Black-Scholes model as you have it, there is of course no need for intermediate steps when pricing vanilla calls, since the SDE has the closed-form solution you included. Intermediate steps would be required for complicated payoffs or other SDEs.

To answer your question though, you do need to use additional dimensions. Think of the option pricing algorithm as an integration over the probability space of stock price paths. Each intermediate tenor introduces a new dimension to that probability space.

The point of these quasirandom sequences is that, in multiple dimensions, they provide more evenly-distributed coverage of the probability space than pseudorandom numbers would give us.


Quasirandom sampling in 2D

If we don't make multidimensional draws from the Sobol sequence, we would not benefit from that extra regularity.


Pseudorandom sampling in 2D

Note how these pseudorandom draws have multiple spots where samples are right next to each other, along with some very big "holes".


You need to use more dimensions. If the number of dimensions (i.e. steps) is large, you may also have to use a Brownian bridge as described in the book by Joshi or Jäckel.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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