nyblom_test {tstests}

Nyblom-Hansen Parameter Constancy Test

Description

The parameter constancy test of Nyblom (1989).

Usage

nyblom_test(
  x,
  scores = NULL,
  parameter_names = colnames(scores),
  parameter_symbols = NULL,
  ...
)

Arguments

x

a series representing the standardized residuals of some estimated model.

scores

the log likelihood score matrix. The estfun method if exported by a package for a model will return this matrix.

parameter_names

optional character vector of the parameter names. Usually read off the column names of the score matrix.

parameter_symbols

an optional character vector of the latex names of the parameters which can be used when printing using the flextable format.

...

not currently used.

Details

The p-values for the test statistic are based on a pre-computed density, by simulation using equation 3.3 of Nyblom (1989), with up to 40 parameters and saved as an internal data object within the package. A kernel density is used to fit the 10,000 samples of the distribution before extracting the p-values. The original simulation generated more than 100,000 data points but these were compressed to quantiles at intervals of 0.001 in order to keep the package size under 5MB.

Value

An object of class “tstest.nyblom” which has a print and as_flextable method.

References

Nyblom,J. (1989). “Testing for the constancy of parameters over time.” Journal of the American Statistical Association, 84(405), 223–230.

Examples

library(tsgarch)
library(xts)
data("spy")
spyr <- na.omit(diff(log(spy)))
spec <- garch_modelspec(spyr[1:1200], model = "garch", order = c(1,1),
constant = TRUE, distribution = "norm")
mod <- estimate(spec)
test <- nyblom_test(residuals(mod, standardize = TRUE), scores = estfun(mod),
parameter_names = names(coef(mod)),
parameter_symbols = mod$parmatrix[estimate == 1]$symbol)
print(test)