When Option Pricing Requires Numerical Methods
Summary
The document asks which options lack closed-form pricing formulas and describes several cases where numerical methods are commonly used. The answer points to American options, whose early-exercise feature requires checking whether exercising is preferable at different points during the contract. Tree-based methods are given as one way to evaluate that choice. Options with nonlinear payoffs or trigger conditions, including binary and knock-in structures, are also identified as candidates for numerical modeling.
A second answer challenges the distinction between analytical and numerical pricing. It argues that if market data include an implied volatility surface, prices can be mapped from observable inputs and contract details, even for more complex products. This raises a definitional point rather than demonstrating a closed-form solution: using a price surface as an input is not the same as deriving a payoff’s value from a model analytically. The document gives examples but no pricing derivations, comparisons, or guidance on choosing a numerical method.
Key ideas
- American options often require evaluating whether early exercise is optimal during the option’s life.
- Tree methods can represent exercise decisions at successive points in time.
- Nonlinear payoffs and trigger conditions can make numerical pricing methods useful.
- An implied volatility surface can map market inputs to prices, but that does not necessarily constitute a closed-form model solution.
Tags
Full text
# Option pricing without analytical solutions # Option pricing without analytical solutions I am quite new to the topic of financial options. I'm aware of options with analytical solutions (e.g. European options in Black-Scholes and Ornstein-Uhlenbeck models). I read that sometimes (most times?) numerical methods are used for option pricing. I also read that there exist exotic options which simply do not have a closed-form solution and require numerical methods; however, I have not been able to find examples of such options. Could anyone name a few? Cheers. ## Answer by D Stanley (score 4) https://quant.stackexchange.com/a/51275 American options (on any underlying) is the first that comes to mind. They are often priced using a tree-based algorithm to determine if there is a benefit to early exercise anywhere along the life of the option. Any options that has non-linearity or trigger conditions (binary, knock-in, etc.) are also candidates for numerical models. ## Answer by will (score -2) https://quant.stackexchange.com/a/51287 You can price anything with an analytic formula if you have all of the required parameters. Let's take european options as an example. We can price these using Black76 if we have the forward price, the appropriate discount factor, expiry, the strike, and the volatility. This last bit - the volatility - needs to have an important consideration: it is the black76 option price implied volatility. If we are pricing using some other model, say Vasicek, then we need a different volatility. That you're given all of the parameters already is the important point - the model is popular enough that you can often get a vol surface (where the volatility is the black76 vol, assuming a specific conventions around stuff like your calculation of T (i.e. act/act 250 business days per year, etc.)) provided as market data. The vol surface is essentialyl a one to one mapping to the option prices (given all the other params). Now, if we have some other, more exotic stucture, there's nothing to stop us doing the same thing - we can construct some function that takes in the market data (i.e. numbers observable directly in the market), specifics of the trade, and then some preconstructed surface which maps to price of the instrument. Does this last example count as analytic? If it does, then anything can be priced analytically...
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.