compute.Hprior: Penalise complexity prior for the Hurst exponent
In eirikmn/INLA.climate: Full Bayesian analysis of radiative forcing climate model

compute.Hprior

R Documentation

Penalise complexity prior for the Hurst exponent

Description

Computes a penalise complexity prior function for the Hurst exponent in logarithmic scaling. Both fGn and ARFIMA(0,d,0) models are supported.

Usage

compute.Hprior(n, upper, alpha, persistent=TRUE, model="fgn")

Arguments

`n`	Length of the time series for which the prior is applied. Increasing this will theoretically improve the accuracy of the prior, but will be more costly to perform. Experiments have suggested that the prior is rather invariant of this parameter.
`upper`	The value of the penalise complexity parameter `upper`.
`alpha`	The value of the penalise complexity parameter `alpha`. This also denotes the probability that the Hurst exponent exceeds `upper`.
`persistent`	A boolean variable. If `TRUE`, the PC-prior will be created for only the persistent range `0.5<H<1`.
`model`	Determines which model the Hurst exponent belongs to. Currently, only fractional Gaussian noise (`model="fgn"`) and ARFIMA(0,d,0) (`model="arfima"`, d=H-0.5) models are supported.

Details

The PC-prior parameters shape the density as the probability of the Hurst exponent exceeding upper is equal to alpha.

Value

Returns the PC-prior in logarithmic scaling as a function.

Author(s)

Eirik Myrvoll-Nilsen eirik.myrvoll-nilsen@uit.no

References

Sørbye, S., Rue, H. (2017) Fractional Gaussian noise: Prior specification and model comparison. Environmetrics, 29(5-6)

Examples

HH = seq(0.501,0.999,by=0.0001)
l.HH = log((HH-0.5)/(1-HH))

l.pcp.f = compute.Hprior(100,0.9,0.01,persistent=TRUE,model="fgn")
pcp = 0.5+0.5/(1+exp(-l.pcp.f(l.HH)))

plot(x=HH,y=pcp,type="l",xlab="H",ylab="Density")

eirikmn/INLA.climate documentation built on Feb. 6, 2023, 11:41 a.m.