bdpopt (version 1.0-1)

optimise.sequential.normal.eu: Optimise A Sequential Normal Decision Problem

Description

Optimise a sequential normal decision problem constructed by a call to the function sequential.normal.dp.

Usage

optimise.sequential.normal.eu(dp, range, step.size,
prior.mean = 0, n.sims = 1000, plot.results = TRUE)

Arguments

dp

A sequential normal decision problem created by a call to sequential.normal.dp.

range

The range of the one-dimensional grid for the state. All grid points will lie in the interval c(prior.mean - range / 2, prior.mean + range / 2).

step.size

The step size between the grid points for the state.

prior.mean

Defines the mid-point of the grid for the state.

n.sims

The number of random draws for simulations used to estimate the expected utility for a decision at each stage.

plot.results

If TRUE, construct a plot of the optimal policy.

Value

A list with components

opt.decision

Function taking a stage and a state into the optimal decision corresponding to the closest grid point at that stage.

opt.utility

Function taking a stage and a state into the optimal expected utility corresponding to the closest grid point at that stage.

Details

The plot has the stage number on the x-axis. The y-axis levels of the points (one for each stage) shows the cutoff levels for when it is optimal to continue (for all but the last stage) or finalise (for the last stage). It will be optimal to continue or finalise if the posterior mean of the parameter at a given stage is above the level of the point at that stage.

If no cutoff point can be established in the interior of the region defined by the interval I = c(prior.mean - range / 2, prior.mean + range / 2), then a cross will be used instead of a circle to indicate the level of the stage. If the cross is located at prior.mean + range / 2, it is optimal to continue for no state in I. If the cross is located at prior.mean - range / 2, it is optimal to continue for all states in I.

See Also

optimise.sequential.eu