Option Pricing Models and Numerical Implementation Methods
Article Quant Q&A · Author: amr
Summary
The document surveys option pricing models and the computational methods used to turn them into prices. Model families include Black–Scholes and related formulas, local and stochastic volatility, Lévy processes, jump models, and fixed income and multi-factor models. It also notes that commodities may need specialized assumptions. The discussion distinguishes the description of asset behavior from the technique used to calculate an option value.
Implementation approaches include closed-form and approximate formulas, binomial or trinomial trees, finite-difference or finite-element solutions to pricing PDEs, Monte Carlo simulation, and interpolation across known option prices. Fourier methods using the fast Fourier transform and risk-neutral valuation are also mentioned in responses. The material is a broad list rather than a comparative study: it provides no benchmarks or systematic criteria for choosing a method, and it flags computational cost and early exercise as practical Monte Carlo concerns.
Key ideas
- Option pricing models specify assumptions about how underlying assets evolve, while numerical methods implement those models.
- Local volatility, stochastic volatility, Lévy processes, and jump models extend the basic lognormal framework in different ways.
- Trees and PDE solvers approximate prices through discretization, while Monte Carlo estimates values by simulating paths.
- Closed-form formulas, moment-matched approximations, Fourier methods, and interpolation provide other routes to prices.
- Monte Carlo can handle many models and payoff types, but accuracy can require substantial computation and early exercise adds complications.
Tags
Full text
# Methods for pricing options # Methods for pricing options I'm looking at doing some research drawing comparisons between various methods of approaching option pricing. I'm aware of the Monte Carlo simulation for option pricing, Black-Scholes, and that dynamic programming has been used too. Are there any key (or classical) methods that I'm missing? Any new innovations? Recommended reading or names would be very much appreciated. Edit: Additionally, what are the standard methods and approaches for assessing/analysing a model or pricing approach? ## Answer by SetTheorist (score 24, accepted) https://quant.stackexchange.com/a/732 There are a wide variety of models (by which I mean the theoretical / mathematical formulation of how the underlying financial variable(s) of interest behave). The most popular ones differ depending on the asset class under consideration (though some are mathematically the same and named differently). Some examples are: - Black-Scholes / Black / Garman-Kohlhagan - Local-volatility [aka Dupire model] - Stochastic-volatility - a generic term for extensions of Black-Scholes where there is a second stochastic factor driving the volatility of the spot; examples are Heston, SABR - Levy processes (usually actually log-Levy): a wide class of models with some features that make them theoretically / technically nice; examples are VG, CGMY - jumps (often compound Poisson) of various kinds can be added to the above models, for example Merton model is Black-Scholes with jumps - CIR, OU processes show up in fixed-income - There are multi-factor (ie multiple driving Brownian motions) versions of the above; e.g. Libor market model, correlated log-normal models - Models for pricing, say, credit-default swaps are often Poisson processes with random hazard rates - Commodities such as electricity can require specialized models to handle the particular features of that market - etc. Implementation methodologies can include: - analytic formulae (usually involving special functions): examples are the classic case of European-style vanilla options in Black-Scholes / CEV / VG; many exotics in black-scholes can be "solved" in this way - approximate analytic - for example, one might price average-rate options in Black-Scholes by approximating the final distribution (by moment-matching) with a shifted-lognormal and using the closed-form for the shifted-lognormal - Binomial / trinomial trees can be viewed as a discretization technique for approximating, say, Black-Scholes. (Note that some people might view the approximation as a model in its own right --- a conflation of model & implementation and more of a philosophical stance than a practical consideration.) - Numerical methods for solving or approximating the PDE governing the option price; this could be solved by finite-difference methods, finite-element methods, etc. - Monte-carlo is a nice brute-force way to handle almost any kind of model and most options (though there are complications with early-exercise style features of options), but it typically takes a lot of computing power to get any accuracy in the price - Interpolation could be viewed as a technique --- if you know the price of a collection of options (varying in some parameters) you can price a new option by interpolating based on the parameters (volatility surfaces implemented by interpolating a grid of given options are examples of this) - etc. ## Answer by Ralph Winters (score 6) https://quant.stackexchange.com/a/695 I would also look into pricing models based upon models other than lognormal (Black-Scholes). Do some research on "fat tailed" or stable distributions. There can also be known by their specific distribution names as Levy, Levy-Poisson, or Cauchy. http://en.wikipedia.org/wiki/Fat_tail ## Answer by nkhuyu (score 5) https://quant.stackexchange.com/a/6992 Fourier Transform seems a good method for option pricing by take advantage of Fast Fourier Transform technique, such as the following paper written by Peter Carr and Dilip B. Madan: http://portal.tugraz.at/portal/page/portal/Files/i5060/files/staff/mueller/FinanzSeminar2012/CarrMadan_OptionValuationUsingtheFastFourierTransform_1999.pdf ## Answer by GKED (score 4) https://quant.stackexchange.com/a/698 There are two more methods that i am aware of from my academic curriculum. I am not sure if they are applicable to the real world but you can read up on them. Method 1: Binomial Valuation; Method 2: Risk Neutral Valuation; Both of the methods are fairly easy to implement(in terms of writing a program for it or simply, using excel spreadsheets) ## Answer by marie albertini (score 0) https://quant.stackexchange.com/a/81421 there is an "other" way to price option based on a mix of pde, tree and monte carlo as described in the article I cited (too long to just put in my answer)...worth looking at, it is based on monotonic interpolation, see https://arxiv.org/pdf/2411.05425
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.