brms (version 0.6.0)

macf: Autocorrelation Function Estimation based on MCMC-Samples (experimental)

Description

Compute (and plot by default) the autocorrelation function based on MCMC samples

Usage

macf(x, lag.max = NULL, plot = TRUE, ...)

Arguments

x

A matrix, data.frame or numeric vector representing the time series. Each row is interpreted as a sample from an MCMC procedure, whereas the columns are taken to be points of a time series. The colnames of x are taken to be the groups and it is assumed

lag.max

Maximum lag at which to calculate the autocorrelation. Default is 10*log10(N/g) where N is the number of observations and g the number of groups. Will be automatically limited to one less than the number of maxim

plot

logical. If TRUE (the default) results are plotted directly

...

Further arguments to be passed to plot.acf.

Value

• An object of class acf.

Examples

## investigate the autorcorrelation in the residuals of a fitted model
## simulate data
set.seed(123)
phi <- c(0.4, 0.7, -0.3)
y <- 0
y[2] <- phi[1] * y[1] + rnorm(1)
y[3] <- phi[1] * y[2] + phi[2] * y[1] + rnorm(1)
for (i in 4:300) y[i] <- sum(phi * y[(i-1):(i-3)]) + rnorm(1)

## fit the model
fit1 <- brm(y ~ 1)
summary(fit1)

## investigate the residuals (autocorrelation clearly visible)
macf(residuals(fit1, summary = FALSE), lag.max = 10)

## fit the model again with autoregressive coefficients
fit2 <- brm(y ~ 1, autocor = cor_ar(p = 3))
summary(fit2)

## investigate the residuals again (autocorrelation is gone)
macf(residuals(fit2, summary = FALSE), lag.max = 10)