Package {gctsc}

gctsc: Copula Count Time Series Models

Description

Provides likelihood-based inference for Gaussian and Student-t copula models for univariate count time series. Supports Poisson, negative binomial, binomial, beta-binomial, and zero-inflated marginals with ARMA dependence structures. Includes simulation, maximum-likelihood estimation, residual diagnostics, and predictive inference. Implements Time Series Minimax Exponential Tilting (TMET) doi:10.1016/j.csda.2026.108344, an adaptation of minimax exponential tilting of Botev (2017) doi:10.1111/rssb.12162. Also provides a linear-cost implementation of the Geweke–Hajivassiliou–Keane (GHK) simulator following Masarotto and Varin (2012) doi:10.1214/12-EJS721, and the Continuous Extension (CE) approximation of Nguyen and De Oliveira (2025) doi:10.1080/02664763.2025.2498502. The package follows the S3 design philosophy of 'gcmr' but is developed independently.

Author(s)

Maintainer: Quynh Nguyen nqnhu2209@gmail.com

Authors:

• Victor De Oliveira

See Also

Useful links:

• https://github.com/QNNHU/gctsc

• Report bugs at https://github.com/QNNHU/gctsc/issues

Daily aggregated weather measurements for KCWC station

Description

Daily aggregated weather measurements for KCWC station

Usage

data("KCWC")

Format

A data frame with 2665 rows and 4 variables.

ARMA Correlation Structure for Copula Count Time Series Models

Description

Constructs an ARMA(p,q) correlation structure for use in Gaussian and Student–t copula count time series models.

Usage

arma.cormat(p = 0, q = 0, tau.lower = NULL, tau.upper = NULL)

Arguments

Details

The ARMA model specifies the dependence structure of the latent copula process.

The ARMA parameters must define a stationary and invertible process. These conditions are enforced during model fitting.

Value

An object of class "arma.gctsc" and "cormat.gctsc" containing:

• npar: Number of ARMA parameters (p + q).

• od: Integer vector c(p, q).

• start: Function to compute starting values from data, typically using arima.

• lower, upper: Parameter bounds.

See Also

gctsc, poisson.marg, predict.gctsc

Weekly Campylobacter case counts across Germany

Description

Weekly Campylobacter case counts across Germany

Usage

data("campyl")

Format

A data frame with 1248 rows and 413 variables.

Extract Coefficients from a gctsc Model

Description

Returns the estimated coefficients from a fitted gctsc model object.

Usage

## S3 method for class 'gctsc'
coef(object, ...)

Arguments

Value

A named numeric vector of model coefficients.

Fit a Copula-Based Count Time Series Model

Description

Fits a Gaussian or Student–t copula model to a univariate count time series with flexible discrete marginal distributions and latent ARMA dependence.

Usage

gctsc(
  formula = NULL,
  data,
  marginal,
  cormat,
  method = c("TMET", "GHK", "CE", "GHK_mvt"),
  c = 0.5,
  QMC = TRUE,
  pm = 30,
  start = NULL,
  family = "gaussian",
  df = 10,
  options = gctsc.opts()
)

Arguments

Details

Supported marginal distributions include Poisson, negative binomial, binomial, beta-binomial, and their zero-inflated variants. The latent dependence is specified through a correlation model such as arma.cormat.

The copula likelihood involves a high-dimensional rectangle probability. This probability is approximated using one of the following methods:

• "TMET": Time Series Minimax Exponential Tilting,

• "GHK": Geweke–Hajivassiliou–Keane simulation,

• "CE": Continuous Extension,

• "GHK_mvt": GHK approximation for the multivariate Student–t rectangle probability. This option is experimental and is mainly intended for comparisons.

The model interface follows the usual R formula convention. An intercept is included by default and can be removed using -1 or 0 +. For non-zero-inflated marginals, formula may be a standard formula, such as y ~ x1 + x2, or a named list list(mu = y ~ x1 + x2). For zero-inflated marginals, formula must be a named list with components mu and pi0, for example list(mu = y ~ x1 + x2, pi0 = ~ z1).

Formula interface. For non-zero-inflated marginals, users may write either formula = y ~ x1 + x2 or formula = list(mu = y ~ x1 + x2). Internally, both are represented as a list with component mu.

For zero-inflated marginals, users must supply both the mean/count component and the zero-inflation component:

formula = list(mu  = y ~ x1 + x2,pi0 = ~ z1 + z2)

The response variable is taken from formula$mu. The pi0 formula should be one-sided. Intercepts are handled by model.matrix following standard R formula rules.

Missing values. Missing values are not handled automatically. Users should remove or impute missing values before calling gctsc. This avoids ambiguity in the time series dependence structure.

Dependence. The dependence parameters are encoded through cormat. ARMA(0,0) is not supported. For ARMA dependence, admissible starting values should satisfy the usual causality and invertibility conditions.

Seed and numerical stability. Simulation-based likelihood approximations are random unless a seed is supplied. If options$seed is provided, the same random stream is used across likelihood evaluations, which can make optimization and standard error estimation more stable. If no seed is supplied, the model can still be fitted, but the approximate likelihood and numerical Hessian may be less stable.

Value

An object of class "gctsc" containing, among others:

• coef: parameter estimates;

• maximum: approximate maximized log-likelihood;

• se: standard errors, when available;

• formula: normalized model formula list;

• terms: model terms for each component;

• model: model frames for each component;

• call: matched function call.

References

Nguyen, Q. N. and De Oliveira, V. (2026), Approximating Gaussian Copula Models for Count Time Series: Connecting the Distributional Transform and a Continuous Extension, Journal of Applied Statistics, 53: 1–22.

Nguyen, Q. N. and De Oliveira, V. (2026), Likelihood Inference in Gaussian Copula Models for Count Time Series via Minimax Exponential Tilting, Computational Statistics & Data Analysis, 218: 108344.

Nguyen, Q. N. and De Oliveira, V. (2026), Scalable Likelihood Inference for Student–t Copula Count Time Series, Stats, 9: 1–49.

See Also

gctsc.opts, arma.cormat, poisson.marg, zip.marg, zib.marg, zibb.marg

Examples

## Example 1: Gaussian copula, Poisson marginal, AR(1)
set.seed(42)
n <- 500
sim_dat <- sim_poisson(mu = 10, tau = 0.3, arma_order = c(1, 0),
                       nsim = n, family = "gaussian")

dat <- data.frame(y = sim_dat$y)

fit_gauss <- gctsc(
  y ~ 1,
  data = dat,
  marginal = poisson.marg(lambda.lower = 0),
  cormat = arma.cormat(p = 1, q = 0), family = "gaussian",
  method = "CE",
  options = gctsc.opts(M = 1000, seed = 42)
)
summary(fit_gauss)

## Example 2: Student--t copula
sim_dat_t <- sim_poisson(mu = 10, tau = 0.3, arma_order = c(1, 0),
                         nsim = 500, family = "t", df = 10)

dat_t <- data.frame(y = sim_dat_t$y)

fit_t <- gctsc(
  y ~ 1,
  data = dat_t,
  marginal = poisson.marg(lambda.lower = 0),
  cormat = arma.cormat(p = 1, q = 0), family ="t",
  df= 10, method = "CE",
  options = gctsc.opts(M = 1000, seed = 42)
)
summary(fit_t)

Worked Examples Included with the Package

Description

The gctsc package includes additional worked examples in the installed ‘examples/’ directory.

Details

Worked examples are installed with the package in the ‘examples/’ directory. They are organized into the subfolders ‘gaussian/’ and ‘student_t/’, corresponding to the two copula families supported by the package.

The top-level example directory can be located with system.file("examples", package = "gctsc").

To list the available Gaussian examples, use

dir(file.path(system.file("examples", package = "gctsc"), "gaussian"))

To list the available Student-t examples, use

dir(file.path(system.file("examples", package = "gctsc"), "student_t"))

To inspect the contents of one of the example scripts without running it, use:

f <- file.path(system.file("examples", package = "gctsc"),
               "gaussian", "poisson.R")
cat(readLines(f), sep = "\n")

Examples

exdir <- system.file("examples", package = "gctsc")
dir(exdir)

Set Options for Gaussian and Student t Copula Time Series Model

Description

Creates a control list for simulation and likelihood approximation in the Gaussian and Student t copula model, including the random seed, Monte Carlo settings, and optimization controls.

Usage

gctsc.opts(seed = 1, M = c(100, 1000), ...)

Arguments

Value

A list with components:

Marginal Model Constructors for gctsc

Description

These functions construct marginal model objects for use with gctsc. Each constructor returns an object of class "marginal.gctsc" containing the information needed to initialize marginal parameters and compute the latent truncation bounds used in the copula likelihood.

The following marginal families are currently supported:

• Poisson: poisson.marg()

• Negative Binomial: negbin.marg()

• Binomial: binom.marg()

• Beta–Binomial: bbinom.marg()

• Zero-Inflated Poisson: zip.marg()

• Zero-Inflated Binomial: zib.marg()

• Zero-Inflated Beta–Binomial: zibb.marg()

Supported link functions depend on the marginal family:

• poisson.marg(), zip.marg(), and negbin.marg() support "identity" and "log".

• binom.marg(), bbinom.marg(), zib.marg(), and zibb.marg() currently support "logit" only.

Usage

poisson.marg(link = "log", lambda.lower = NULL, lambda.upper = NULL)

binom.marg(link = "logit", size = NULL, lambda.lower = NULL, lambda.upper = NULL)

zib.marg(link = "logit", size = NULL, lambda.lower = NULL, lambda.upper = NULL)

negbin.marg(link = "log", lambda.lower = NULL, lambda.upper = NULL)

zip.marg(link = "log", lambda.lower = NULL, lambda.upper = NULL)

bbinom.marg(link = "logit", size, lambda.lower = NULL, lambda.upper = NULL)

zibb.marg(link = "logit", size,  lambda.lower = NULL, lambda.upper = NULL)

Arguments

Details

Marginal Models for Copula Count Time Series

The marginal constructors are designed to be supplied to gctsc, rather than called during likelihood evaluation by the user. Each returned marginal object contains internal functions for:

• computing starting values for the marginal parameters;

• determining the number of marginal parameters;

• converting observed counts into lower and upper latent truncation bounds for the copula likelihood.

Internally, the design input x is represented as a named list of design matrices. For non-zero-inflated marginals, x contains x$mu. For zero-inflated marginals, x contains both x$mu and x$pi0. These matrices are constructed automatically by gctsc from the supplied formula and data.

For zero-inflated marginals, the mu component controls the main count distribution, while the pi0 component controls the structural zero probability through a logit link.

The optional bounds lambda.lower and lambda.upper are attached to the starting values and used during numerical optimization.

Value

A marginal model object of class "marginal.gctsc".

See Also

gctsc, arma.cormat

Examples

poisson.marg(link = "log")
negbin.marg(link = "log")
binom.marg(link = "logit", size = 10)
bbinom.marg(link = "logit", size = 24)
zip.marg(link = "log")
zib.marg(link = "logit", size = 10)
zibb.marg(link = "logit", size = 24)

Diagnostic Plots for Fitted Copula Count Time Series Models

Description

Produces diagnostic plots for a fitted Gaussian or Student–t copula count time series model of class "gctsc".

The diagnostics are based on randomized quantile residuals and probability integral transform (PIT) values.

Usage

## S3 method for class 'gctsc'
plot(
  x,
  caption = rep("", 5),
  main = rep("", 5),
  level = 0.95,
  col.lines = "gray",
  ...
)

Arguments

Details

The following diagnostic plots are produced:

1. Time series of randomized quantile residuals.

2. Q–Q plot against the reference distribution.

3. Histogram of PIT values.

4. Autocorrelation function (ACF) of residuals.

5. Partial autocorrelation function (PACF) of residuals.

For Gaussian copulas, residuals are compared against the standard normal distribution. For Student–t copulas, residuals are compared against a Student–t distribution with degrees of freedom obtained from fitted model.

Value

Invisibly returns NULL.

Invisibly returns NULL. The function is called for its side effect of producing diagnostic plots.

See Also

residuals.gctsc

residuals.gctsc for computing the residuals used in the plots.

Examples

# Simulate data from a Poisson AR(1) model
set.seed(123)
n <- 2000
mu <- 5
phi <- 0.5
arma_order <- c(1, 0)
y <- sim_poisson(mu = mu, tau = phi, arma_order = arma_order, nsim = n)$y

# Fit the model using the CE method
fit <- gctsc(y~1, data = data.frame(y),
  marginal = poisson.marg(link = "identity", lambda.lower = 0),
  cormat = arma.cormat(p = 1, q = 0), family ="gaussian",
  method = "CE",
  options = gctsc.opts(seed = 1, M = 1000),
  c = 0.5
)

# Produce diagnostic plots
par(mfrow = c(2, 3))
plot(fit)

Approximate Gaussian Copula Log-Likelihood

Description

Computes an approximate log-likelihood for a Gaussian copula count time series model from latent lower and upper truncation bounds. The approximation method is selected through the argument method.

Usage

pmvn(
  lower,
  upper,
  tau,
  od,
  method = c("CE", "GHK", "TMET"),
  c = 0.5,
  pm = 30,
  M = 1000,
  QMC = TRUE,
  ret_llk = TRUE
)

Arguments

Details

The package currently supports three likelihood approximations: continuous extension (CE), Geweke–Hajivassiliou–Keane simulation (GHK), and Time Series Minimax Exponential Tilting (TMET).

The function pmvn() provides a unified interface for Gaussian copula likelihood approximation. The argument method selects among:

• "CE": continuous extension approximation,

• "GHK": sequential importance sampling via the GHK simulator,

• "TMET": minimax exponential tilting approximation.

The arguments c, pm, M, and QMC are used only by the methods to which they apply.

Value

A numeric scalar giving the approximate log-likelihood. If ret_llk = FALSE, method-specific diagnostic output is returned.

References

Nguyen, Q. N. and De Oliveira, V. (2026). Approximating Gaussian Copula Models for Count Time Series: Connecting the Distributional Transform and a Continuous Extension, Journal of Applied Statistics, 53, 1–22.

Nguyen, Q. N., and De Oliveira, V. (2026), Likelihood Inference in Gaussian Copula Models for Count Time Series via Minimax Exponential Tilting, Computational Statistics and Data Analysis, 218: 108344.

See Also

pmvt, sim_poisson, poisson.marg

Examples

mu <- 10
tau <- 0.2
arma_order <- c(1, 0)

sim_data <- sim_poisson(mu = mu, tau = tau, arma_order = arma_order,
                        nsim = 500, family = "gaussian", seed = 1)
y <- sim_data$y

lower <- qnorm(ppois(y - 1, lambda = mu))
upper <- qnorm(ppois(y, lambda = mu))

## Continuous extension
pmvn(lower, upper, tau = tau, od = arma_order, method = "CE", c = 0.5)

## GHK approximation
pmvn(lower, upper, tau = tau, od = arma_order, method = "GHK", M = 1000)

## TMET approximation
pmvn(lower, upper, tau = tau, od = arma_order, method = "TMET",
     pm = 30, M = 1000)

Approximate Student–t Copula Log-Likelihood

Description

Computes an approximate log-likelihood for a Student–t copula count time series model from latent lower and upper truncation bounds. The approximation method is selected through the argument method.

Usage

pmvt(
  lower,
  upper,
  tau,
  od,
  method = c("CE", "GHK", "TMET"),
  c = 0.5,
  pm = 30,
  M = 1000,
  QMC = TRUE,
  ret_llk = TRUE,
  df
)

Arguments

Details

The package currently supports three likelihood approximations: continuous extension (CE), Geweke–Hajivassiliou–Keane simulation (GHK), and Time Series Minimax Exponential Tilting (TMET).

The function pmvt() provides a unified interface for Student–t copula likelihood approximation. The argument method selects among:

• "CE": continuous extension approximation,

• "GHK": sequential importance sampling via the GHK simulator,

• "TMET": minimax exponential tilting approximation.

The arguments c, pm, M, and QMC are used only by the methods to which they apply.

Value

A numeric scalar giving the approximate log-likelihood. If ret_llk = FALSE, method-specific diagnostic output is returned.

References

Nguyen, Q. N. and De Oliveira, V. (2026). Scalable Likelihood Inference for Student–t Copula Count Time Series, Stats, 9: 1–49.

See Also

pmvn, gctsc, sim_poisson

Examples

mu <- 10
tau <- 0.2
arma_order <- c(1, 0)
df <- 8

sim_data <- sim_poisson(mu = mu, tau = tau, arma_order = arma_order,
                        nsim = 200, family = "t", df = df, seed = 1)
y <- sim_data$y

lower <- qt(ppois(y - 1, lambda = mu), df = df)
upper <- qt(ppois(y, lambda = mu), df = df)

## Continuous extension
pmvt(lower, upper, tau = tau, od = arma_order, method = "CE", c = 0.5, df = df)

## GHK approximation
pmvt(lower, upper, tau = tau, od = arma_order, method = "GHK", M = 200, df = df)

## TMET approximation
pmvt(lower, upper, tau = tau, od = arma_order, method = "TMET", pm = 30, M = 200, df = df)

One-Step-Ahead Predictive Distribution for Copula Count Time Series Models

Description

Computes the one-step-ahead predictive distribution for a fitted Gaussian or Student–t copula count time series model.

The predictive probability mass function is evaluated over the grid 0:y_max. If y_max is not supplied, it is chosen automatically from the fitted response values as ceiling(max(y) + k * sd(y)). Summary statistics of the predictive distribution are returned. If the observed response is included in newdata, the Continuous Ranked Probability Score (CRPS) and Logarithmic Score (LOGS) are also computed.

Usage

## S3 method for class 'gctsc'
predict(object, newdata = NULL, y_max = NULL, k = 3, ...)

Arguments

Details

The function constructs prediction design matrices from the stored model terms using model.matrix. Thus the same formula convention used in model fitting is used for prediction, including automatic intercept handling and factor-variable expansion.

For zero-inflated marginals, newdata is used to construct both the mean-component design matrix and the zero-inflation design matrix. The column names of the new design matrices must match those from the fitted model.

The predictive distribution is computed using the copula family and approximation method stored in the fitted object. For Gaussian copulas, multivariate normal rectangle probabilities are used. For Student–t copulas, multivariate t rectangle probabilities are used.

If the observed response is included in newdata, it must be a single nonnegative integer count and should not exceed y_max.

Value

A list containing:

• mean: Predictive mean.

• median: Predictive median.

• mode: Predictive mode.

• variance: Predictive variance.

• p_y: Predictive probability mass function evaluated on y_grid.

• y_grid: Count grid 0:y_max over which the predictive distribution is evaluated.

• lower, upper: Lower and upper endpoints of the 95\ predictive interval.

• CRPS: Continuous Ranked Probability Score, returned only if the observed response is supplied in newdata.

• LOGS: Logarithmic Score, returned only if the observed response is supplied in newdata.

• y_true: Observed response value, returned only if supplied in newdata.

References

Nguyen, Q. N. and De Oliveira, V. (2026), Likelihood Inference in Gaussian Copula Models for Count Time Series via Minimax Exponential Tilting, Computational Statistics & Data Analysis, 218: 108344.

Nguyen, Q. N. and De Oliveira, V. (2026), Scalable Likelihood Inference for Student–t Copula Count Time Series, Stats, 9: 1–49.

See Also

gctsc, arma.cormat, gctsc.opts

Examples

# Simulate Poisson AR(1) data
set.seed(1)
y_sim <- sim_poisson(
  mu = 10,
  tau = 0.2,
  arma_order = c(1, 0),
  nsim = 200,
  family = "gaussian"
)$y

dat <- data.frame(y = y_sim)

# Fit Gaussian copula model
fit <- gctsc(
  formula = y ~ 1,
  data = dat,
  marginal = poisson.marg(link = "log"),
  cormat = arma.cormat(p = 1, q = 0),
  method = "GHK",
  family = "gaussian",
  options = gctsc.opts(M = 1000, seed = 42)
)

# One-step-ahead prediction for an intercept-only model
pred <- predict(fit, y_max = 30)

# If the future observed value is available, include it in newdata
pred_score <- predict(fit, newdata = data.frame(y = 8), y_max = 30)

Print a gctsc Model Object

Description

Displays the call, estimation method, parameter estimates, and likelihood information.

Usage

## S3 method for class 'gctsc'
print(x, digits = max(3, getOption("digits") - 3), ...)

Arguments

Print Summary of a gctsc Model

Description

Displays summary statistics and model fit information for a fitted gctsc model.

Usage

## S3 method for class 'summary.gctsc'
print(x, digits = 4, ...)

Arguments

Randomized Quantile Residuals for Copula Count Time Series Models

Description

Computes randomized quantile residuals for a fitted Gaussian or Student–t copula count time series model.

For discrete responses, residuals are constructed using the randomized probability integral transform (PIT) of Dunn and Smyth (1996). The conditional probabilities required for the PIT are approximated according to the fitted copula family and likelihood method. For Gaussian copula models fitted by TMET or GHK, the same simulation-based method is used in the residual computation. For Gaussian copula models fitted by CE, the required conditional probabilities are approximated by GHK. For Student–t copula models, the required conditional probabilities are approximated by GHK.

Usage

## S3 method for class 'gctsc'
residuals(object, ...)

Arguments

Details

For observation y_t, let F_t(y_t^-|y) and F_t(y_t|y) denote the conditional CDF evaluated at y_t - 1 and y_t given past observations, respectively. The PIT value is computed as

e_t = F_t(y_t^-|y) + u_t \{F_t(y_t|y) - F_t(y_t^-|y)\},

where u_t \sim \mathrm{Uniform}(0,1).

For Gaussian copulas, residuals are obtained as r_t = \Phi^{-1}(e_t).

For Student–t copulas with degrees of freedom df, the residuals are defined as r_t = t_{\nu}^{-1}(e_t), where t_{\nu}^{-1} denotes the quantile function of the Student–t distribution.

Value

A list of class "gctsc.residuals" containing:

• residuals: Numeric vector of randomized quantile residuals.

• pit: Numeric vector of probability integral transform values.

References

Nguyen, Q. N. and De Oliveira, V. (2026), Approximating Gaussian Copula Models for Count Time Series: Connecting the Distributional Transform and a Continuous Extension, Journal of Applied Statistics, 53: 1–22.

Nguyen, Q. N. and De Oliveira, V. (2026), Likelihood Inference in Gaussian Copula Models for Count Time Series via Minimax Exponential Tilting, Computational Statistics & Data Analysis, 218: 108344.

Nguyen, Q. N. and De Oliveira, V. (2026), Scalable Likelihood Inference for Student–t Copula Count Time Series, Stats, 9: 1–49.

See Also

gctsc, sim_gctsc, pmvn, pmvt, predict.gctsc

Examples

# Simulate Poisson AR(1) data under a Gaussian copula
set.seed(1)
y <- sim_poisson(mu = 5, tau = 0.7,
                 arma_order = c(1, 0),
                 nsim = 500,
                 family = "gaussian")$y

fit <- gctsc(
  y ~ 1,
  data = data.frame(y),
  marginal = poisson.marg(),
  cormat = arma.cormat(1, 0),
  family = "gaussian",
  method = "CE",
  options = gctsc.opts(seed = 1, M = 1000)
)

res <- residuals(fit)
hist(res$residuals, main = "Randomized Quantile Residuals")
hist(res$pit, main = "PIT Histogram")

Weekly Rotavirus case counts across Germany

Description

Weekly Rotavirus case counts across Germany

Usage

data("rota")

Format

A data frame with 1248 rows and 413 variables.

Simulate from Gaussian and t Copula Time Series Models

Description

These functions simulate time series data from Gaussian and t copula models with various discrete marginals and an ARMA dependence structure.

Usage

sim_poisson(
  mu,
  tau,
  arma_order,
  nsim,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_negbin(
  mu,
  dispersion,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_zip(
  mu,
  pi0,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_binom(
  prob,
  size,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_bbinom(
  prob,
  rho,
  size,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_zib(
  prob,
  pi0,
  size,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

sim_zibb(
  prob,
  rho,
  pi0,
  size,
  tau,
  arma_order,
  nsim = 100,
  family = c("gaussian", "t"),
  df = NULL,
  seed = NULL
)

Arguments

Details

Marginal types:

• Poisson: Counts with mean \mu.

• Negative binomial (NB): Overdispersed counts with mean \mu and dispersion parameter \kappa.

• Binomial: Number of successes in n trials with success probability p.

• Beta–-binomial (BB): Binomial with success probability p following a beta distribution, allowing intra-cluster correlation \rho.

• Zero–inflated Poisson (ZIP): Poisson with extra probability \pi_0 of an excess zero.

• Zero–inflated binomial (ZIB): Binomial with extra probability \pi_0 of an excess zero.

• Zero–inflated beta–binomial (ZIBB): Beta–binomial with extra probability \pi_0 of an excess zero.

Parameterization notes:

• Negative binomial uses dispersion (\kappa) to model overdispersion: larger dispersion increases variance for a fixed mean.

• Beta–binomial and ZIBB use rho as the overdispersion parameter: \rho is the intra-class correlation, with \rho \to 0 giving the binomial model.

• Zero–inflated marginals include a separate pi0 parameter that controls the extra probability mass at zero.

Worked examples. Additional worked examples, including Gaussian and Student–t copula models with zero-inflated marginals, are provided in the installed example scripts; see gctsc-examples.

Value

A list with components:

• y: Simulated time series data.

• z: Latent Gaussian process values.

• marginal: Marginal distribution name.

• parameters: List of parameters used.

• cormat: ARMA structure.

See Also

gctsc, sim_gctsc, marginal.gctsc, pmvn, pmvt, predict.gctsc

Examples

# Poisson example
sim_poisson(mu = 10, tau = c(0.2, 0.2), 
  arma_order = c(1, 1), nsim = 100, 
  family = "gaussian", seed = 42)

# Negative Binomial example
sim_negbin(mu = 10, dispersion = 2, tau = c(0.5, 0.5),
  arma_order = c(1, 1),family = "gaussian",
  nsim = 100, seed =1)

# Zero Inflated Beta-Binomial example with seasonal covariates
n <- 100
xi <- numeric(n)
zeta <- rnorm(n)
for (j in 3:n) {
  xi[j] <- 0.6 * xi[j - 1] - 0.4 * xi[j - 2] + zeta[j]
}
prob <- plogis(0.2 + 0.3 * sin(2 * pi * (1:n) / 12) +
             0.5 * cos(2 * pi * (1:n) / 12) + 0.3 * xi)
sim_zibb(prob, rho = 1/6, pi0 = 0.2, size = 24, tau = 0.5,
 arma_order = c(1, 0),family = "t", df = 10, nsim = 100)

Summarize a gctsc Model Fit

Description

Computes standard errors, z-values, and p-values for the estimated parameters in a fitted gctsc object.

Usage

## S3 method for class 'gctsc'
summary(object, ...)

Arguments

Value

A list of class summary.gctsc containing model summary statistics.