Skip to content
All library documents

Choosing Binomial Lattice Resolution Through Convergence Analysis

Article Quant Q&A · Author: Kch
Read the original

Summary

The document asks how many time levels to use when valuing a European call on a zero-coupon bond in a Black-Derman-Toy interest-rate lattice. It frames the choice as a balance: a coarse lattice may lack precision, while a fine lattice increases computational cost. The response gives a rough starting point, suggesting that about one hundred levels may be a minimum, while stressing that the appropriate resolution depends on available computing resources.

The practical recommendation is to measure accuracy against runtime across candidate resolutions before putting a numerical algorithm into production. This convergence analysis helps select a lattice size based on the particular valuation and implementation. The response does not provide a convergence table, error tolerance, or details about the option and model parameters, so its suggested minimum should not be treated as a universal rule. The number of levels must be justified by results for the specific calculation.

Key ideas

  • Lattice resolution trades valuation precision against computational cost.
  • The response suggests roughly one hundred levels as a possible minimum, depending on computing capacity.
  • Quants can select parameters by measuring numerical accuracy against runtime across resolutions.
  • The document gives no error thresholds or convergence results, so its suggested level count is only a starting point.

Tags

Fixed incomeOptionsDerivatives pricingBacktesting
Full text
# Optimal number of nodes for binomial lattice?


# Optimal number of nodes for binomial lattice?












Let's suppose one is valuing a Euro call on a ZCB in a Black-Derman-Toy lattice. How many nodes/levels of discretization are optimal? Obviously too many creates computational issues and too few creates low precision. What is the best way to determine the optimal number of nodes?

## Answer by Antoine Conze (score 2)

https://quant.stackexchange.com/a/39986

100 is probably a minimum but it depends on your computational power at hand. Usually before putting a numerical algorithm in production quants run extensive convergence accuracy vs. CPU time analysis to decide on the appropriate parameters setup.

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.