forward.sim.policy: Forward simulation based on a sequence of emulators

In mludkov/mlOSP: Machine Learning and Regression Monte Carlo Algorithms for Optimal Stopping

Forward simulation based on a sequence of emulators

Description

Simulate h(X_tau) using FIT (can be a dynaTree, smooth.spline, MARS or RandomForest or RVM or hetGP)

Usage

forward.sim.policy(
  x,
  M,
  fit,
  model,
  offset = 1,
  compact = TRUE,
  use.qv = FALSE
)

Arguments

x	is a matrix of starting values if input x is a list, then use the grids specified by x
M	number of time steps to forward simulate
fit	a list of fitted emulators that determine the stopping classifiers to be used
model	a list containing all model parameters
offset	(internal for debugging purposes)
compact	flag; if FALSE returns additional information about forward x-values.
use.qv	boolean to indicate whether to plug-in continuation value for allpaths still alive at the last time-step. Default is set to FALSE

Details

Should be used in conjuction with the osp.xxx functions that build the emulators. Also called internally from osp.fixed.design

Value

a list containing:

• payoff is the resulting payoff NPV from t=0

• fvalue[i] is a list of resulting payoffs (on paths still not stopped) NPV from t=i

• tau are the times when stopped

• sims is a list; sims[[i]] are the forward x-values of paths at t=i (those not stopped yet) nsims number of total 1-step simulations performed

mludkov/mlOSP documentation built on April 29, 2023, 7:56 p.m.