R/distr_skellam_meanvar.R

In gasmodel: Generalized Autoregressive Score Models

# SKELLAM DISTRIBUTION / MEAN-VARIANCE PARAMETRIZATION


# Parameters Function ----------------------------------------------------------
distr_skellam_meanvar_parameters <- function(n) {
  group_of_par_names <- c("mean", "var")
  par_names <- c("mean", "var")
  par_support <- c("real", "positive")
  res_parameters <- list(group_of_par_names = group_of_par_names, par_names = par_names, par_support = par_support)
  return(res_parameters)
}
# ------------------------------------------------------------------------------


# Density Function -------------------------------------------------------------
distr_skellam_meanvar_density <- function(y, f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  res_density <- be_silent(exp(-s) * ((s + m) / (s - m))^(y / 2) * besselI(x = sqrt(s^2 - m^2), nu = y))
  res_density[!is.finite(res_density)] <- -Inf
  return(res_density)
}
# ------------------------------------------------------------------------------


# Log-Likelihood Function ------------------------------------------------------
distr_skellam_meanvar_loglik <- function(y, f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  res_loglik <- be_silent(y / 2 * log((s + m) / (s - m)) - s + log(besselI(x = sqrt(s^2 - m^2), nu = y)))
  res_loglik[!is.finite(res_loglik)] <- -Inf
  return(res_loglik)
}
# ------------------------------------------------------------------------------


# Mean Function ----------------------------------------------------------------
distr_skellam_meanvar_mean <- function(f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  res_mean <- m
  return(res_mean)
}
# ------------------------------------------------------------------------------


# Variance Function ------------------------------------------------------------
distr_skellam_meanvar_var <- function(f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  res_var <- s
  res_var <- array(res_var, dim = c(t, 1, 1))
  return(res_var)
}
# ------------------------------------------------------------------------------


# Score Function ---------------------------------------------------------------
distr_skellam_meanvar_score <- function(y, f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  tri_bi_y <- (besselI(x = sqrt(s^2 - m^2), nu = y - 1) + besselI(x = sqrt(s^2 - m^2), nu = y + 1)) / besselI(x = sqrt(s^2 - m^2), nu = y)
  res_score <- matrix(0, nrow = t, ncol = 2L)
  res_score[, 1] <- (s * y) / (s^2 - m^2) - m / (2 * sqrt(s^2 - m^2)) * tri_bi_y
  res_score[, 2] <- -(m * y) / (s^2 - m^2) + s / (2 * sqrt(s^2 - m^2)) * tri_bi_y - 1
  return(res_score)
}
# ------------------------------------------------------------------------------


# Fisher Information Function --------------------------------------------------
distr_skellam_meanvar_fisher <- function(f) {
  t <- nrow(f)
  m <- f[, 1, drop = FALSE]
  s <- f[, 2, drop = FALSE]
  m[s <= abs(m)] <- NA_real_
  s[s <= abs(m)] <- NA_real_
  tri_bi_m <- (besselI(x = sqrt(s^2 - m^2), nu = m - 1) + besselI(x = sqrt(s^2 - m^2), nu = m + 1)) / besselI(x = sqrt(s^2 - m^2), nu = m)
  res_fisher <- array(0, dim = c(t, 2L, 2L))
  res_fisher[, 1, 1] <- m^2 / (4 * (s^2 - m^2)) * ((2 * s) / (sqrt(s^2 - m^2)) - tri_bi_m)^2
  res_fisher[, 1, 2] <- -(m * s) / (4 * (s^2 - m^2)) * ((2 * s) / (sqrt(s^2 - m^2)) - tri_bi_m)^2
  res_fisher[, 2, 1] <- res_fisher[, 1, 2]
  res_fisher[, 2, 2] <- s^2 / (4 * (s^2 - m^2)) * ((2 * s) / (sqrt(s^2 - m^2)) - tri_bi_m)^2
  return(res_fisher)
}
# ------------------------------------------------------------------------------


# Random Generation Function ---------------------------------------------------
distr_skellam_meanvar_random <- function(t, f) {
  m <- f[1]
  s <- f[2]
  if (s > abs(m)) {
    res_random <- be_silent(stats::rpois(t, lambda = (s + m) / 2) - stats::rpois(t, lambda = (s - m) / 2))
  } else {
    res_random <- rep(NA_real_, times = t)
  }
  res_random <- matrix(res_random, nrow = t, ncol = 1L)
  return(res_random)
}
# ------------------------------------------------------------------------------


# Starting Estimates Function --------------------------------------------------
distr_skellam_meanvar_start <- function(y) {
  y_mean <- mean(y, na.rm = TRUE)
  y_var <- stats::var(y, na.rm = TRUE)
  m <- y_mean
  s <- max(y_var, abs(y_mean) + 1e-6)
  res_start <- c(m, s)
  return(res_start)
}
# ------------------------------------------------------------------------------