monte carlo simulation vs finite difference method

Discussion in 'Options' started by rkdlbh, Oct 19, 2015.

  1. rkdlbh

    rkdlbh

    hello all

    Probably a very basic question for most but for someone just starting to learn about options, has proven quite problematic!

    Currently working on a project regarding these two option price predicting techniques. really struggling to work out how they each work. I kind of understand that Monte Carlo is about expectations. You run a bunch of simulations using a RNG, transform that random number into an outcome for your simulation...

    but really struggling on how finite difference method works. please could someone just provide me with a basic overview on how this work and which method is better to use? thanks in advance
     
    lawrence-lugar likes this.
  2. You will most likely get better answers to these type of questions on either Wilmott or Nuclearphynance.
     
  3. rkdlbh

    rkdlbh

    thanks. will try them!
     
  4. xandman

    xandman

    Don't shoo away a potential topic that could up our readership. I have met a ton of rocket scientists on this board alone. Most are just too quiet.

    Here is the old college try:

    Monte Carlo simulation is an approximation via use of random numbers. It is commonly used by the Statistical branch of the mathematical sciences.

    Here is a simple example of using Monte Carlo to calculate Pi. It is usually introduced in beginning comp sci. It has an interesting history , too.

    https://learntofish.wordpress.com/2010/10/13/calculating-pi-with-the-monte-carlo-method/

    In the Applied Maths branch, you have Finite Difference methods which is a numerical method (an efficient technique unlike the way you solve by hand) of solving partial differential equations (PDE). That's part of Calculus for you heathens.

    For FDM, I leave you to research the use on wikipedia. As I don't know anything more about those topics :p
     
    Last edited: Oct 19, 2015
  5. nitro

    nitro

    In a word, finite differences is about estimating derivatives. Monte Carlo is about estimating integrals. The relationship is the relationship between derivative and integrals in the form of the

    Fundamental Theorem of Calculus:

    ftoc.gif

    The second fundamental theorem of calculus holds for [​IMG] a continuous function on an open interval [​IMG] and [​IMG] any point in [​IMG], and states that if [​IMG] is defined by

    ne2.gif

    and now you see how MC and FDs are related by using this equation (integrals and derivatives are mathematical inverses of each other):

    ne3.gif

    Sometimes it is better to approach a problem through derivatives, sometimes through integrals. The FTOC and SFTOC tells you that you can recover the solution from either end. But just like when you are taking a trip, one way to the destination might be considerably harder than another - not just computationally, but theoretically.
     
    Last edited: Oct 19, 2015
    curiousGeorge8 and xandman like this.
  6. xandman

    xandman

    I never thought they were that closely related. Good stuff.
     
  7. xandman

    xandman


    Oops. I re-read your post. This isn't one of those "Why is the sky blue?" questions. Your really do need the guys at Wilmott or nuc phynance. o_O
     
  8. That is basic calculus nothing crazy.
     
  9. xandman

    xandman



    Hint: Duffy Financial Instruments in C++. Chapter 19 has "My First Finite Difference Solver" with UML and code.
     
    Last edited: Oct 19, 2015
    JefeTrader likes this.
  10. IAS_LLC

    IAS_LLC

    Sorry if I'm being repetitive (I didn't read the full thread chain)

    Finite differences and Monte Carlo Methods are not mutually exclusive. They can and SHOULD be used jointly.

    Finite differences are used to to solve systems of partial differential equations (PDEs) numerically. If those PDEs contain any sort of stochastic component (noise) , than one should use Monte Carlo Simulation to vary the stochastic components of those PDEs and see how it affects the solution/trajectories the PDEs take. If you are doing options pricing using Black-Scholes, than you want to solve the Black Scholes equation (a PDE) and your monte carlo variables could be the stocks price, the interest rate, the drift rate, the standard deviation of returns, etc... Performing the Monte Carlo simulation will allow you to establish the Expected Value of the contract at some future date, as well as the distribution of that contracts price given stochastic variation of the parameters of the Black-Scholes model
     
    #10     Oct 20, 2015
    ironchef likes this.