pred.density: Predictive Densities for bma Objects

Description

Predictive densities for conditional forecasts

Usage

pred.density(object, newdata = NULL, n = 300, hnbsteps = 30, ...)

Value

pred.density returns a list of class pred.density with the following elements

densities(): a list whose elements each contain the estimated density for each forecasted observation
fit: a vector with the expected values of the predictions (the 'point forecasts')
std.err: a vector with the standard deviations of the predictions (the 'standard errors')
dyf(realized.y, predict_index=NULL): Returns the densities of realized response variables provided in realized.y.
If realized.y is a matrix, then each row corresponds to a forecast observation in newdata
if not left empty, predict.index specifies to which observations in newdata the realized.y should apply
lps(realized.y, predict_index=NULL): Computes the log predictive score for the response varaible provided in realized.y (cf. lps.bma) -
Note that the LPS equals minus the mean of the logarithmized results from dyf)
plot((x, predict_index = NULL, addons = "eslz", realized.y = NULL, addons.lwd = 1.5, ...): the same as plot.pred.density
n: The number of equally spaced points for which the density (under densities() was computed.
nmodel: The number of best models predictive densities are based upon.
call: the call that created this pred.density object

Arguments

object: a bma object - see bms, alternativel a zlm object
newdata: A data.frame, matrix or vector containing variables with which to predict.
n: The integer number of equally spaced points at which the density is to be estimated.
hnbsteps: The number of numerical integration steps to be used in case of a hyper-g prior (cf. argument g in bms). Increase this number to increase accuracy. Must be an even integer.
...: arguments to be passed on to plot.density.

Details

The predictive density is a mixture density based on the nmodels best models in a bma object (cf. nmodel in bms).
The number of 'best models' to retain is therefore vital and should be set quite high for accuracy.

Examples

Run this code


 data(datafls)
 mm=bms(datafls,user.int=FALSE)
 
 #predictive densityfor two 'new' data points
 pd=pred.density(mm,newdata=datafls[1:2,]) 
 
 
 #fitted values based on best models, same as predict(mm, exact=TRUE)
 pd$fit
 
 #plot the density for the first forecast observation
 plot(pd,1)  
 
 # the same plot ' naked'
 plot(pd$densities()[[1]])
 
 
 #predict density for the first forecast observation if the dep. variable is 0
 pd$dyf(0,1) 
 
 #predict densities for both forecasts for the realizations 0 and 0.5
 pd$dyf(rbind(c(0,.5),c(0,.5)))
 
 # calc. Log Predictive Score if both forecasts are realized at 0:
 lps.bma(pd,c(0,0))