R: Predictive Densities for bma Objects

pred.density {BMS}

R Documentation

Predictive Densities for bma Objects

Description

Predictive densities for conditional forecasts

Usage

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

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.

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

Note

In BMS version 0.3.0, pred.density may only cope with built-in gpriors, not with any user-defined priors.

Author(s)

Martin Feldkircher and Stefan Zeugner

Examples

 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))

[Package BMS version 0.3.4 Index]