# Finite-Difference Option Pricing: Grids, Stability, and Early Exercise

Understand how explicit, implicit, and Crank–Nicolson grids solve an option-pricing PDE, enforce American exercise, and require boundary, convergence, and model checks.

Canonical: https://wiki.fcontext.com/options/finite-difference-option-pricing/
Fact checked: 2026-07-22

> For educational purposes only; not investment advice.

<a id="answer"></a>

## Direct answer

**Finite-difference option pricing** replaces a continuous pricing equation with a grid of underlying-price and time points, approximates derivatives by differences between neighboring nodes, and works backward from the terminal payoff to today's theoretical value. It is useful when no convenient closed form exists, especially for American exercise, barriers, time-dependent inputs, and other boundary-dependent contracts.

The method does not make a price objective or tradable. Its output depends on the stochastic model, contract payoff, dividends, volatility, rates, grid limits, boundary conditions, numerical scheme, and convergence tolerance. A polished decimal from an unconverged grid is still a wrong answer.

<a id="mechanism"></a>

## From the pricing PDE to a grid

Under constant-parameter Black–Scholes assumptions with continuous dividend yield `q`, value `V(S,t)` satisfies:

`∂V/∂t + 0.5σ²S² ∂²V/∂S² + (r−q)S ∂V/∂S − rV = 0`.

Choose price nodes `S_i=iΔS` up to `Smax` and time nodes separated by `Δt`. At maturity set the payoff, such as `max(K−S,0)` for a Put. Approximate Delta with neighboring values and Gamma with the centered second difference, then step backward subject to low- and high-price boundary conditions.

- **Explicit:** computes the prior layer directly from the known layer. It is simple but conditionally stable; an overly large `Δt` relative to `ΔS` can produce oscillations or impossible prices.
- **Implicit:** solves a tridiagonal linear system each step. It is generally more stable, but introduces more time discretization error and computational work than the explicit update.
- **Crank–Nicolson:** averages the spatial operator across two time layers. It is usually second-order in time for smooth solutions, but the payoff kink or a discontinuous feature can cause oscillations; initial implicit steps or Rannacher smoothing are common remedies.

For an American option, every node must satisfy:

`V(S,t) ≥ immediate exercise payoff`.

After estimating continuation value, take the larger of continuation and intrinsic value, or solve the equivalent linear-complementarity problem. The boundary where the choice switches is the free exercise boundary.

<a id="example"></a>

## American Put grid and convergence record

Consider an American Put with `S=$95`, `K=$100`, `T=0.5` year, `σ=25%`, `r=4%`, and `q=0`. Let `Smax=$200`, `ΔS=$1`, and `Δt=0.0005` year. At maturity, the node at `$95` starts at intrinsic value `$5`.

Working backward, suppose a step produces continuation value `$6.18` at one near-maturity `$95` node; the solver retains `$6.18` rather than immediate exercise value `$5`. At a `$70` node, if continuation is `$29.72` while immediate exercise is `$30`, the American constraint retains `$30`. Repeating this process builds both the value surface and an approximate early-exercise boundary.

| Price nodes | Time steps | Illustrative value at `$95` |
| ---: | ---: | ---: |
| `100` | `250` | `$8.64` |
| `200` | `1,000` | `$8.68` |
| `400` | `4,000` | `$8.69` |

These figures illustrate a convergence log, not a universal fair value. Compare the European version with Black–Scholes, compare the American result with a deep binomial tree or independent solver, and refine the grid until price and Greeks meet stated tolerances.

<a id="risks"></a>

## Model implementation checklist

- Encode the exact payoff, exercise style, dividends, barriers, observations, settlement, and corporate-action terms.
- State the stochastic process and whether volatility, rates, dividends, or borrow costs vary by price and time.
- Place `Smax` far enough away and test sensitivity; a convenient boundary can contaminate value near spot.
- Derive boundary conditions for the actual contract rather than reusing vanilla conditions blindly.
- Refine `ΔS` and `Δt` together; log price, Delta, Gamma, and exercise-boundary convergence.
- Check explicit stability restrictions and inspect for negative values, oscillations, or violated monotonicity.
- Treat payoff kinks, barriers, discrete dividends, and local-volatility interpolation carefully.
- Use an appropriate American solver and report its residual tolerance.
- Benchmark special cases against formulas, no-arbitrage bounds, trees, or independent code.
- Separate numerical, parameter, and model error; grid refinement fixes only the first.

Market validation still requires synchronized Bid/Ask quotes. A model midpoint can differ from an executable price because of skew, supply and demand, hedge costs, fees, liquidity, discrete trading, jumps, and model risk.

<a id="misconceptions"></a>

## Common misconceptions

- “A finer grid guarantees a correct price.” It can converge precisely to the wrong model or payoff.
- “Crank–Nicolson is always stable and accurate.” Stability does not prevent oscillation, boundary error, or bad inputs.
- “American value is European value plus a fixed premium.” The optimal exercise boundary changes with state and time.
- “The PDE handles jumps automatically.” A diffusion PDE omits jumps unless the operator explicitly includes them.
- “Matching one market quote validates the solver.” Calibration can hide numerical or structural error.
- “Theoretical value is an execution signal.” It does not guarantee a Bid, Ask, hedge, or arbitrage.

<a id="related"></a>

## Related topics

- [Black-Scholes model](/options/black-scholes-model/)
- [Binomial option pricing](/options/binomial-option-pricing/)
- [Early exercise](/options/early-exercise/)
- [Dupire local volatility](/options/dupire-local-volatility/)

<a id="sources"></a>

## Primary sources

- [Black and Scholes (1973): The Pricing of Options and Corporate Liabilities](https://doi.org/10.1086/260062)
- [Brennan and Schwartz (1977): The Valuation of American Put Options](https://doi.org/10.1111/j.1540-6261.1977.tb03284.x)
- [Crank and Nicolson (1947): A Practical Method for Numerical Evaluation of PDE Solutions](https://doi.org/10.1007/BF02127704)
- [Options Industry Council: Black-Scholes Formula](https://www.optionseducation.org/advancedconcepts/black-scholes-formula)
- [Options Industry Council: Exercising Options](https://www.optionseducation.org/optionsoverview/exercising-options)