qgarch_fit: Fit generalized qgarch(m, n) models

Description

Fits QGARCH-in-mean models using nonlinear minimization of the negative log-likelihood. Four variants are supported: a zero-lambda model, a restricted-lambda model, a free-lambda model, and a threshold model with state-dependent lambda.

Usage

qgarch_fit(
  x,
  model = c("zero", "restricted", "free", "threshold"),
  arch_order = 1L,
  garch_order = 1L,
  threshold_indicator = NULL,
  start = NULL,
  rho = 1,
  steptol = 1e-10,
  typsize = 0.1,
  iterlim = 300L,
  print.level = 0,
  hessian = TRUE,
  vcov_type = c("auto", "sandwich", "hessian", "opg", "none")
)

Value

An object of class "qgarch".

Arguments

x: A numeric vector or one-column data frame containing the observed series. The series must contain only finite, non-missing values.
model: Character string specifying the model variant. Must be one of "zero", "restricted", "free", or "threshold".
arch_order: Positive integer giving the ARCH lag order m.
garch_order: Positive integer giving the GARCH lag order n.
threshold_indicator: Optional threshold indicator used only when model = "threshold". Must have the same length as x. Nonzero values are converted to 1 and zero values to 0.
start: Optional starting values. May be NULL, a numeric vector, or a list of numeric vectors.
rho: Scalar used in the restricted-lambda mapping.
steptol: Step tolerance passed to stats::nlm().
typsize: Typical size passed to stats::nlm(). A single value is repeated to the appropriate length.
iterlim: Maximum number of iterations passed to stats::nlm().
print.level: Print level passed to stats::nlm().
hessian: Logical; should the Hessian be returned by stats::nlm()?
vcov_type: Character string controlling standard-error estimation. One of "auto", "sandwich", "hessian", "opg", or "none". In "auto" mode, Hessian-based standard errors are tried first.