Learn R Programming

MultiATSM (version 1.5.0)

Transition_Matrix: Computes the transition matrix required in the estimation of the GVAR model

Description

Computes the transition matrix required in the estimation of the GVAR model

Usage

Transition_Matrix(t_First, t_Last, Economies, type, DataConnectedness)

Value

matrix or list of matrices. Time-varying or time-invariant transition matrix depending on 'type'.

Arguments

t_First

character. Sample starting date (format: yyyy).

t_Last

character. Sample ending date (format: yyyy).

Economies

character vector. Names of the C economies included in the system.

type

character. Method for computing interdependence. Possible options:

  • "Time-varying": Computes time-varying interdependence and returns weight matrices for each year.

  • "Sample Mean": Returns a single weight matrix with average weights over the sample period.

  • Specific year (e.g., "1998", "2005"): Computes time-invariant interdependence for the specified year.

DataConnectedness

list or data frame. Data used to compute the transition matrix (e.g., trade flows).

Details

If there is missing data for any country in a particular year, the transition matrix will include only NAs.

Examples

Run this code
t_First <- "2006"
t_Last <- "2019"
Economies <- c("China", "Brazil", "Mexico", "Uruguay")
type <- "Sample Mean"
# Load data if Connectedness data from excel, otherwise use pre-saved data
GetExcelData <- FALSE

if (GetExcelData) {
  if (!requireNamespace("readxl", quietly = TRUE)) {
    stop(
      "Please install package \"readxl\" to use this feature.",
      call. = FALSE
    )
    DataPath <- system.file("extdata", "TradeData.xlsx", package = "MultiATSM")
    tab_names_Trade <- readxl::excel_sheets(DataPath)
    list_all_Trade <- suppressMessages(lapply(tab_names_Trade, function(x) {
      readxl::read_excel(path = DataPath, sheet = x)
    }))
    names(list_all_Trade) <- tab_names_Trade

    L <- length(list_all_Trade)

    for (i in 1:L) {
      Countries <- list_all_Trade[[i]][[1]]
      list_all_Trade[[i]] <- as.data.frame(list_all_Trade[[i]][, -1])
      rownames(list_all_Trade[[i]]) <- Countries
    }

    DataConnectedness <- list_all_Trade
  }
} else {
  data(TradeFlows)
  DataConnectedness <- TradeFlows
}

W_mat <- Transition_Matrix(t_First, t_Last, Economies, type, DataConnectedness)

Run the code above in your browser using DataLab