 |
Download a pdf copy of this help file here |
|
Download a pdf copy of this help file here |
See also: Monte Carlo simulation introduction, Monte Carlo sampling, Other sampling methods
Latin Hypercube sampling, or LHS, is an option that is now available for most simple risk analysis simulation software programs. It uses a technique known as 'stratified sampling without replacement' (Iman et al., 1980). The probability distribution is split into n intervals of equal probability, where n is the number of samples that are to be performed on the model. As the simulation progresses each of the n intervals is sampled once.
LHS has the advantage of generating a set of samples that more precisely reflect the shape of a sampled distribution than pure random (Monte Carlo) samples. The general effect is that the mean of a set of simulation results more quickly approaches the ‘true’ value, particularly for models that are simply adding or subtracting a number of variables.
ModelRisk includes a number of multivariate distributions, as well as time series, copulas and other multivariate functions none of which are compatible with LHS. This is because LHS can be applied to a single random variable but not over a joint distribution.
In reality, LHS has a limited impact on the model output's accuracy the more distributions there are in a model since LHS only applies to distributions individually. The benefit of LHS is also eroded if one does not complete the number of samples nominated at the beginning, i.e. if one halts the simulation run in mid-simulation.
LHS also applies a heavy burden on a simulation model with a large number
of inputs because it needs to generate and organize samples from each
distribution prior to running the first sample from a distribution. This
can cause a long delay in running a large model, yet provides very little
additional accuracy.
Read on: Other sampling methods