Method comparison
One contract, every method. European options are measured against the closed form; American options against a Leisen-Reimer tree with over two thousand steps, since no closed form exists.
Contract
$
$
Numerical settings
$10.4506
| Method | Price | Error | Relative | Time |
|---|---|---|---|---|
| Black-Scholes reference | 10.4506 | 0.0000 |
0.000% | 0.0 ms |
| Binomial (Cox-Ross-Rubinstein) | 10.4466 | -0.0040 |
-0.038% | 58.2 ms |
| Binomial (Jarrow-Rudd) | 10.4534 | +0.0028 |
+0.027% | 19.3 ms |
| Binomial (Tian) | 10.4472 | -0.0034 |
-0.032% | 19.8 ms |
| Binomial (Leisen-Reimer) | 10.4506 | 0.0000 |
0.000% | 20.9 ms |
| Trinomial | 10.4466 | -0.0040 |
-0.038% | 26.5 ms |
| Finite difference (Crank-Nicolson) | 10.4407 | -0.0099 |
-0.095% | 52.0 ms |
| Monte Carlo | 10.4662± 0.0085 | +0.0157 |
+0.150% | 323.2 ms |
| Barone-Adesi-Whaley | American options only | |||
| Bjerksund-Stensland | American options only | |||
Greeks by method
| Method | Delta | Gamma | Vega | Theta | Rho | Epsilon |
|---|---|---|---|---|---|---|
| Black-Scholes | 0.63683 | 0.018762 | 37.524 | -6.414 | 53.232 | -63.683 |
| Binomial (Cox-Ross-Rubinstein) | 0.63677 | 0.018794 | 37.505 | -6.4202 | 53.23 | -63.677 |
| Binomial (Jarrow-Rudd) | 0.63679 | 0.018772 | 37.745 | -6.4063 | 52.637 | -63.09 |
| Binomial (Tian) | 0.63699 | 0.018777 | 37.427 | -6.4824 | 54.793 | -65.241 |
| Binomial (Leisen-Reimer) | 0.63674 | 0.018784 | 37.524 | -6.414 | 53.232 | -63.683 |
| Trinomial | 0.63665 | 0.018794 | 37.505 | -6.4202 | 53.23 | -63.677 |
| Finite difference (Crank-Nicolson) | 0.63658 | 0.018791 | 37.578 | -6.4244 | 53.226 | -63.667 |
| Monte Carlo | 0.63645 | 0.018753 | 37.558 | -6.4208 | 53.182 | -63.648 |
Absolute error
Distance from the reference price, in currency units.
Compute time
Milliseconds for price and all Greeks, on this server.
When to use which
| Method | Strengths | Weaknesses | Best for |
|---|---|---|---|
| Black-Scholes | Exact and instant. Analytic Greeks. The benchmark every numerical method is judged against. |
European exercise only. Assumes constant volatility and rates and a lognormal underlying. |
Vanilla European options; implied volatility; sanity-checking numerical methods. |
| Binomial (Cox-Ross-Rubinstein) | Simple to explain and implement. Handles American exercise naturally. The tree recombines around the spot. |
Converges slowly, order 1/N, and the price oscillates as the strike moves between nodes. |
Teaching the mechanics of backward induction and early exercise. |
| Binomial (Jarrow-Rudd) | Equal probabilities keep the arithmetic simple. Never produces negative probabilities. |
Same slow, oscillating convergence as Cox-Ross-Rubinstein. The tree drifts, so the spot is not a node after step zero. |
Illustrating that many trees converge to the same limit. |
| Binomial (Tian) | Matches the first three moments of the lognormal distribution at every step. |
No faster than Cox-Ross-Rubinstein in practice; still oscillates. |
Comparing moment-matching against the classic scheme. |
| Binomial (Leisen-Reimer) | Places a node on the strike, so convergence is smooth and of order 1/N². Accurate to a few basis points with a few dozen steps. |
Needs an odd step count and the closed-form d1 and d2, so it is tied to the lognormal model. |
Fast, accurate American pricing; the practical choice among trees. |
| Trinomial | Three branches give a better fit per step than two. Equivalent to an explicit finite-difference scheme. Theta comes straight from the tree. |
More nodes per step than a binomial tree, so each step costs more. Still order 1/N convergence. |
Bridging trees and finite-difference methods. |
| Finite difference (Crank-Nicolson) | Solves the pricing PDE on a grid, giving values for every spot and time at once. Handles American exercise and barriers. |
Grid design matters: explicit schemes can be unstable, and the boundary conditions must be chosen with care. |
American options, barriers, and any single-factor model with a known PDE. |
| Monte Carlo | Handles path-dependent payoffs and many underlyings. Error is measurable through the standard error. |
Slow convergence, order 1/√N. Early exercise needs regression techniques. Greeks are noisy. |
Exotic and multi-asset options. |
| Barone-Adesi-Whaley | Almost as fast as the closed form. Gives the early-exercise boundary explicitly. |
An approximation: error grows for long maturities and very high volatility. American only. |
Quick American pricing when a few cents of accuracy is enough, such as scanning many strikes. |
| Bjerksund-Stensland | Fully closed form, no root finding at all, so it is the fastest American method. |
Assumes a flat exercise boundary, so it is biased low. The 2002 two-step version is more accurate. |
Real-time American quotes and as a control variate for Monte Carlo. |