estimate.tsmarch: Estimates a model given a specification.

In tsmarch: Multivariate ARCH Models

Estimates a model given a specification.

Description

Method for estimating one of the models in the package given a specification object.

Usage

## S3 method for class 'cgarch.spec'
estimate(
  object,
  solver = "solnp",
  control = list(trace = 0),
  return_hessian = TRUE,
  ...
)

## S3 method for class 'dcc.spec'
estimate(
  object,
  solver = "solnp",
  control = list(trace = 0),
  return_hessian = TRUE,
  ...
)

## S3 method for class 'gogarch.spec'
estimate(object, trace = FALSE, ...)

Arguments

object	an object of class “cgarch.spec”, “dcc.spec” or “gogarch.spec”.
solver	the solver to use for the second stage estimation of the multivariate dynamics. Valid choices are solnp, nloptr and optim, with the latter using the “L-BFGS-B” method. This option is inactive for the GOGARCH model which uses the default solver in package “tsgarch” for the estimation of the independent factors.
control	solver control parameters.
return_hessian	whether to calculate and return the partitioned hessian of the model (see details). Not applicable in the case of the GOGARCH model.
...	for the GOGARCH model, additional options passed to the radical function.
trace	whether to print tracing information for the GOGARCH model estimation.

Details

DCC and Copula Models

Estimation assumes a 2-stage approach whereby the pre-estimated GARCH models (first stage) are used to estimate the Copula or DCC dynamics. In the case of the constant correlation Gaussian model, there are no parameters to estimate. Whilst this 2-stage approach results in a very fast estimation, the calculation of the standard errors based on the partitioned hessian is quite expensive. The user can create a futures plan to take advantage of parallel functionality which for large dimensional problems leads to a large speedup. Additionally, the option of not calculating the hessian (“return_hessian”) is available. In that case, the scores are still calculated and the resulting standard errors will be based on the outer product of gradients.

GOGARCH Model

The independent factors are calculated by first pre-whitening the data (PCA), selecting the number of factors, and then solving for the rotation matrix. A GARCH model is then estimated on each factor. A minimal amount of information is retained in the estimation object and most of the heavy lifting related to co-moment matrices, weighted moments etc is done through dedicated methods. The estimation method makes use of parallel processing for the independent factor GARCH models which can be setup using plan. Additionally, the RADICAL algorithm benefits from part parallelization which can be controlled using setThreadOptions.

Value

An estimated object of one of either “cgarch.estimate”, “dcc.estimate” or “gogarch.estimate”.

Author(s)

Alexios Galanos

tsmarch documentation built on April 3, 2025, 7:40 p.m.