Systematic Scenario Selection for Stress Testing

Quasi-Monte-Carlo algorithm for systematic generation of shock scenarios from an arbitrary multivariate elliptical distribution. The algorithm selects a systematic mesh of arbitrary fineness that approximately evenly covers an isoprobability ellipsoid in d dimensions (Flood, Mark D. & Korenko, George G. (2013) ). This package is the 'R' analogy to the 'Matlab' code published by Flood & Korenko in above-mentioned paper.

SyScSelection

Quasi-Monte-Carlo algorithm for systematic generation of shock scenarios from an arbitrary multivariate elliptical distribution. The algorithm selects a systematic mesh of arbitrary fineness that approximately evenly covers an isoprobability ellipsoid in d dimensions.
(Flood, Mark D. & Korenko, George G. "Systematic Scenario Selection", Office of Financial Research Working Paper #0005, 2013)

Installation:

install.packages("devtools")
library(devtools)
install_github("mvk222/SyScSelection")
library(SyScSelection)

Usage:

Example ellipsodial mesh for a normal distribution:

• Estimate the mean and covariance matrix from the data:
  mu <- colMeans(data)
sig <- cov(data)

• The number of dimensions, d, is taken directly from the data:
  d <- length(data[1,])

• Get the size parameter for a normal dist’n at a 95% threshold:
  calpha <- sizeparam_normal_distn(.95, d)

• Create a hyperellipsoid object. Note that the constructor takes the inverse of the disperion matrix:
  hellip <- hyperellipsoid(mu, solve(sig), calpha)

• Scenarios are calculated as a mesh of fineness 3. The number of scenarios is a function of the dimensionality of the hyperellipsoid and the fineness of the mesh:
  scenarios <- hypercube_mesh(3, hellip)

Example ellipsodial mesh for a t distribution:

• Estimate the mean, covariance, and degrees of freedom from the data:
  mu <- colMeans(data)
sig <- cov(data)
nu <- dim(data)[1] - 1

• The number of dimensions, d, is taken directly from the data:
  d <- length(data[1,])

• Get the size parameter for a normal dist’n at a 95% threshold:
  calpha <- sizeparam_t_distn(.95, d, nu)

• Create a hyperellipsoid object. Note that the constructor takes the inverse of the disperion matrix:
  hellip <- hyperellipsoid(mu, solve(sig), calpha)

• Scenarios are calculated as a mesh of fineness 3. The number of scenarios is a function of the dimensionality of the hyperellipsoid and the fineness of the mesh:
  scenarios <- hypercube_mesh(3, hellip)