Skip to contents

Create a Posterior for a GEV Model

GEVBayes0.Rd

Create a "Poor Man's" Posterior for a GEV model using MCMC iterates.

Usage

GEVBayes0(MCMC, blockDuration = 1.0,
          MAP = NULL,
          yMax = NULL,
          nMax = length(yMax))

Arguments

MCMC

An object that can be coerced into a matrix containing the MCMC iterates. It should have the burnin period removed and be thinned if necessary.

blockDuration

The block duration given as a single positive numeric value. The GEV distribution which parameters are sampled in MCMC refers to the maximum on a period with duration blockDuration.

MAP

An optional vector of Maximum A Posteriori for the parameter vector. Should be named with names matching the colnames of MCMC.

yMax

An optional vector of observations.

nMax

An optional number of observations. Useful only when yMax is not given.

Value

An object with class "GEVBayes0" inheriting from

"Bayes0". This object can be used to produce RL plots.

Note

The argument yMax is intended for the classical framework where block maxima are used corresponding to a constant block duration. This is equivalent to using the potData argument with the value

potData(MAX.data = as.list(yMax), MAX.effDuration = rep(blockDuration, length(yMax)).

See also

RL method to generate a data frame of "classical" return levels (as shown on a classical RL plot), predict.GEVBayes0 to generate a data frame of predictive return levels (as shown on a predictive RL plot).

Examples

require(revdbayes)
## ========================================================================
## Portpirie data. Note that 'yMax' is only used for graphics later
## ========================================================================
prior <- set_prior(prior = "flatflat", model = "gev")
post <- rpost_rcpp(n = 10000, model = "gev", prior = prior,
                       data = portpirie)
## retrieve the MAP within the object
MAP <- post$f_mode
names(MAP) <- c("loc", "scale", "shape")

postGEV0 <- GEVBayes0(MCMC = post$sim_vals, yMax = portpirie, MAP = MAP)

## ========================================================================
## some methods
## ========================================================================
summary(postGEV0)
#> GEV Model Bayesian Inference
#> o Block duration : 1
#> o Number of blocks used: 65
#> o Number of MCMC iterates: 10000
#> o Posterior mean [sd]:
#>            loc          scale          shape 
#>  3.875 [0.029]  0.207 [0.022] -0.034 [0.100] 
coef(postGEV0)
#>             loc     scale       shape
#> mean   3.874539 0.2068635 -0.03386931
#> median 3.874232 0.2052087 -0.04113732
#> mode   3.874750 0.1980440 -0.05010991
vcov(postGEV0)
#>                 loc         scale        shape
#> loc    0.0008364864  0.0002086140 -0.001008695
#> scale  0.0002086140  0.0004817316 -0.000719470
#> shape -0.0010086955 -0.0007194700  0.010072483

## ========================================================================
## RL plot
## ========================================================================
RL0 <- RL(postGEV0)
autoplot(postGEV0) + ggtitle("GEV fit to Portpirie data")


## ========================================================================
## predictive distribution
## ========================================================================
pred <- predict(postGEV0)
autoplot(pred) 

autoplot(predict(postGEV0, newDuration = 100)) +
    ggtitle("Prediction for a 'new' period of 100 years")