midasml (version 0.0.6)

get_start_midas: MIDAS regression function for initial values

Description

Computes initial values for different MIDAS regression model specifications.

Usage

get_start_midas(
  y,
  X,
  z = NULL,
  loss = c("mse", "als", "rq"),
  weight,
  polynomial,
  num.evals = 1000,
  num.coef = 10,
  seed = NULL,
  tau = tau
)

Arguments

X

predictor variable lags in MIDAS data format.

z

autoregressive lags of response variable (default NULL, DL-MIDAS model is computed).

loss

polynomial specification.

weight

MIDAS weight function (depends on polynomial).

polynomial

MIDAS polynomial specification.

num.evals

number of objective function evaluations using random starting parameter values.

num.coef

number of best coefficients to use as starting values in nonlinear optimization.

seed

value used in set.seed for randomly drawing initial starting values.

tau

quantile level for als and rq regressions.

params

parameter vector from midas_estimate.

Value

returns num.coef number of initial parameter values.

Details

For a given loss function and MIDAS weight function specification, computes num.evals number of random initial values and evaluates the objective function at those parameters. Function retains num.coef of best in terms of fit initial starting values, which are then feed into optimization algorithms.