bayesianOU (version 0.1.3)

fit_ou_nonlinear_tmg: Fit Bayesian nonlinear OU model with TMG effect and SV

Description

Fits a Bayesian nonlinear Ornstein-Uhlenbeck model with cubic drift, stochastic volatility, and Student-t innovations using Stan.

Usage

fit_ou_nonlinear_tmg(
  results_robust,
  Y,
  X,
  TMG,
  COM,
  CAPITAL_TOTAL,
  model = c("base"),
  priors = list(sigma_delta = 0.002),
  com_in_mean = TRUE,
  chains = 6,
  iter = 12000,
  warmup = 6000,
  thin = 2,
  cores = max(1, parallel::detectCores() - 1),
  threads_per_chain = 2,
  hard_sum_zero = TRUE,
  orthogonalize_tmg = TRUE,
  factor_from = c("X", "Y"),
  use_train_loadings = FALSE,
  adapt_delta = 0.97,
  max_treedepth = 12,
  seed = 1234,
  init = NULL,
  moment_match = NULL,
  verbose = FALSE
)

Value

List containing:

factor_ou

Model results including draws and parameter estimates

beta_tmg

Time-varying beta estimates

sv

Stochastic volatility summaries

nonlinear

Nonlinearity diagnostics

accounting

TMG accounting block

diagnostics

MCMC diagnostics, LOO, and OOS metrics

Arguments

results_robust

List. Previous results object to extend (can be empty list).

Y

Numeric matrix (T x S). Dependent variable (prices/values by sector).

X

Numeric matrix (T x S). Independent variable (production prices).

TMG

Numeric vector (length T). Aggregate TMG series.

COM

Numeric matrix (T x S). Composition of capital by sector.

CAPITAL_TOTAL

Numeric matrix (T x S). Total capital by sector.

model

Character. Model type. Currently only "base" supported.

priors

List. Prior specifications. Currently supports sigma_delta.

com_in_mean

Logical. Include COM effect in mean equation. Default TRUE.

chains

Integer. Number of MCMC chains. Default 6.

iter

Integer. Total iterations per chain. Default 12000.

warmup

Integer. Warmup iterations. Default 6000.

thin

Integer. Thinning interval. Default 2.

cores

Integer. Number of cores for parallel chains.

threads_per_chain

Integer. Threads per chain for within-chain parallelism.

hard_sum_zero

Logical. If TRUE, TMG wedge is fixed at zero. Default TRUE.

orthogonalize_tmg

Logical. Orthogonalize TMG w.r.t. common factor. Default TRUE.

factor_from

Character. Source for common factor: "X" or "Y". Default "X".

use_train_loadings

Logical. Compute factor loadings from training only. Default FALSE.

adapt_delta

Numeric. Target acceptance rate (0-1). Default 0.97.

max_treedepth

Integer. Maximum tree depth for NUTS. Default 12.

seed

Integer. Random seed for reproducibility.

init

Numeric or function. Initial values for parameters.

moment_match

Logical. Use moment matching for LOO. Default NULL.

verbose

Logical. Print progress messages. Default FALSE.

Details

The model uses hierarchical priors for sector-specific parameters. Training period is set to 70 percent of observations by default. All data standardization uses training period statistics only.

Examples

# \donttest{
# 1. Prepare dummy data
T_obs <- 20
S_sectors <- 2
Y <- matrix(rnorm(T_obs * S_sectors), nrow = T_obs, ncol = S_sectors)
X <- matrix(rnorm(T_obs * S_sectors), nrow = T_obs, ncol = S_sectors)
TMG <- rnorm(T_obs)
COM <- matrix(runif(T_obs * S_sectors), nrow = T_obs, ncol = S_sectors)
K <- matrix(runif(T_obs * S_sectors, 100, 1000), nrow = T_obs, ncol = S_sectors)

# 2. Run model (conditional on Stan backend availability)
# We use very short chains just to demonstrate execution
if (requireNamespace("cmdstanr", quietly = TRUE) || 
    requireNamespace("rstan", quietly = TRUE)) {
  
  # Wrap in try to avoid failure if Stan is not configured locally
  try({
    results <- fit_ou_nonlinear_tmg(
      results_robust = list(),
      Y = Y, X = X, TMG = TMG, COM = COM, CAPITAL_TOTAL = K,
      chains = 1, iter = 100, warmup = 50, # Short run for example
      verbose = FALSE
    )
  }, silent = TRUE)
}
# }