Create a local GMM model object

Defines a vector-moment model for use with gmm_dnn(). A model supplies observation-level moments and, optionally, observation-level Jacobians and target-specific starting values.

Usage

new_gmm_model(
  name,
  setup,
  moment_obs,
  jacobian_obs = NULL,
  start = NULL,
  coef_names,
  moment_names = NULL
)

Arguments

name

Non-empty model name.

setup

Function that validates and prepares model data.

moment_obs

Function returning an n x q numeric moment matrix.

jacobian_obs

Optional function returning observation-level Jacobians. For scalar theta, an n x q matrix is accepted; otherwise use an n x q x p array.

start

Optional function returning a numeric starting vector.

coef_names

Non-empty character vector naming the parameter entries.

moment_names

Optional character vector naming the moment components.

Value

An object of class gmm_model.

Details

The model functions are called with named arguments: setup(sample, n, ...), moment_obs(sample, theta, ctx), jacobian_obs(sample, theta, ctx), and start(sample, ctx, w, x0). The setup() return value is passed back as ctx.

Examples

mod <- new_gmm_model(
  name = "mean_zero",
  setup = function(sample, n, ...) list(g = sample$g),
  moment_obs = function(sample, theta, ctx) matrix(ctx$g - theta[1], ncol = 1),
  jacobian_obs = function(sample, theta, ctx) matrix(-1, length(ctx$g), 1),
  start = function(sample, ctx, w, x0) sum(w * ctx$g),
  coef_names = "theta",
  moment_names = "mean"
)
is_gmm_model(mod)
#> [1] TRUE